1. The dialect divisions of Middle English are complex and controversial. Divisions based on phonology recognize three to five major dialect areas. In this paper, however, we will be concerned only to show that there was at least one northern dialect and one southern dialect with the characteristics that we describe. Roughly, the two syntactic dialects at issue were found in the North and in the (North)east Midlands, on the one hand, and the South and (South)west Midlands on the other. Within these areas further distinctions can be made that are beyond the scope of this paper.
2. Vikner 1990 calls the IP-V2 languages ``generalized verb-second'' languages because the two he considers, Yiddish and Icelandic, are said to exhibit V2 word order in all types of main and subordinate clauses rather than in the more limited set of environments where it is found in German, Dutch and Mainland Scandinavian. This terminology has the advantage of theoretical neutrality; but, as we shall see, it is inaccurate. IP-V2 languages do not allow V2 word order as freely in subordinate clauses as in main clauses (see also Kemenade (this volume)).
3. Our paper does not take account of recent proposals by Kayne 1994, Zwart 1993, and Roberts (this volume) that treat OV languages as underlyingly VO. If that proposal proves viable, the analyses presented here should be straightforwardly translatable into the new framework.
4. For further discussion of the notion of competition between grammars see Kroch 1989 b, 1994, Pintzuk 1991, Santorini 1992, Taylor 1990, 1994.
5. See, however, Heycock & Kroch 1994 for a more nuanced analysis of V2 sentences with subjects in topic position.
6. Other narrative sequencing adverbs (for example, `nu' ``now'') sometimes behave like `tha', and sometimes like ordinary adverbs.
7. This statement is not entirely uncontroversial. See Diesing 1990.
8. Examples similar to those found in Old English are apparently found in all older West Germanic dialects. Medieval German (Ebert 1986, Behaghel 1932: p. 15) appears to have been intermediate between Old English and modern German in its tolerance for this kind of adjunction. Further work on the V2 syntax of the medieval Germanic languages is needed to determine the proper analysis of these cases.
9. The differences between modern German and the older Germanic languages may be exaggerated by differences in the conventions of the written language at different times. Jack Hoeksema has pointed out to us that the modern German and Dutch counterparts of (9b) are perfectly acceptable with a pause after the initial adverb:
Without the comma as a indicator of the pause, verb-second order is obligatory in the written language. In medieval texts punctuation was much less regular than now, so the absence of commas in (9) does not mean that there were not obligatory pauses after the sentence-initial adverbs.
10. The account in Kemenade 1987, which takes Old English to be a CP-V2 language, also fails to unify Old English clitics with the general Germanic pattern.
11. Though we do not have the space to enter into the matter here, our analysis of the IP-V2 phenomenon has as one of its consequences that main clauses in Icelandic and Yiddish are most likely CP-V2 structures, with IP-V2 limited to non-CP-recursion subordinate clauses.
12. It is worth noting that empty expletive incorporation has as one of its virtues that it provides a mechanism for the agreement relation between finite verbs and nominative objects in Icelandic, a phenomenon not easily treated if all agreement relations are checked in Spec-head configurations.
13. We thank Beatrice Santorini for these examples.
14. Beatrice Santorini points out to us that one interesting feature of our analysis here is that it relates topicalization in modern English more closely to the Old English construction than is usual. In modern English, where V2 does not obtain, the order topic > subject > verb is the only one allowed, and one might ask what licenses the topic position. Our analysis gives an obvious answer: verb-movement to C-zero at LF, just as in Old English. The difference between the two languages is simply that modern English has lost the V2 requirement, perhaps because it lost expletive incorporation, forcing agreement and case to be checked with an overt subject in Spec,IP. The difference between the two languages proposed here would most plausibly have arisen because modern English lost empty expletive incorporation in connection with its loss of empty expletives.
15. The Penn-Helsinki Parsed Corpus of Middle English is a syntactically annotated and somewhat extended version of the prose Middle English section of the Helsinki Corpus of English Texts originally assembled under the direction of Matti Rissanen at the University of Helsinki (see Kytö 1993). The annotation work was done under the direction of Anthony Kroch at the University of Pennsylvania with the support of the National Science Foundation (Grant BNS89-19701) and with supplementary support from the University of Pennsylvania Research Foundation. The annotation scheme was designed by Anthony Kroch and Ann Taylor and implemented by Taylor. The PPCME is available to scholars without fee for educational and research purposes via anonymous ftp from babel.ling.upenn.edu and over the World-Wide Web (http://ling.upenn.edu).
16. The works of Chaucer show a higher rate of inversion in topicalized sentences than other well-known late 14th century texts (Kemenade 1987, Kroch 1989 a). Chaucer was also as likely to invert a pronominal subject with the tensed verb as a non-pronominal one, unlike other authors. If the argument presented below linking inversion with pronoun subjects to Scandinavian influence is correct, Chaucer's syntax may be of a piece with his East Midlands phonology, since the East Midlands were part of the Danelaw. His language may, therefore, indicate a certain conservative regionalism compared to the developing London standard.
17. The discussion in this section is based on an exhaustive sample of the Benet text, which has been entered in its entirety into the PPCME.
18. Constructions similar to one we have found in Benet are not hard
to find in the Germanic dialects. Thus, in modern Dutch such sentences are
found as exclamatives:
The singular article with plural import in (iib) is characteristic of exclamatives. We thank Jack Hoeksema for drawing our attention to these cases.
19. Santorini gives reasons to modify Fanselow's analysis, but in a way that does not affect our reasoning here.
20. We thank Harm Pinkster for bringing this possibility to our attention.
21. We thank Donald Ringe for checking our examples against the Latin and Old English texts in Logeman's (1888) edition of St. Benedict's Rule.
22. Examples like (19) are almost nonexistent in the southern text samples in our corpus. We have found only three, of which two are from the ``Ayenbite of Inwit'' and so are quite late.
23. It might seem odd that Norse speakers should fail to acquire the word-final // of Old English, since their native language contained the sound. In support of our proposal, however, are two facts. First, the distribution of the voiced and voiceless allophones of the phoneme differed in the two languages. Norse had the voiced allophone everywhere but word-initially, while Old English had only the voiceless allophone in word-final position (Noreen 1923, Brunner 1965). Thus, speakers of Norse apparently heard final // as the phonetically similar /s/ because in their language /s/ but not // could occur in word-final position. Furthermore, /s/ in Norse was always voiceless. Second, verbal endings in Old English must have been weakly articulated, hence perceptually unsalient and prone to being misheard by non-native speakers. Evidence for the phonetic weakness of the endings appears in the phonologically unmotivated syncope of the vowel in the endings, though this syncope is characteristic of the southern (West Saxon and Kentish) dialects of Old English and occurs only rarely in Mercian and Northumbrian (Brunner 1965). We thank Donald Ringe for guidance through the philological literature on the points made here.
24. From the usage patterns in Orm and Chaucer (see above), it seems that, besides arising in the North, the CP-V2 grammar also took hold to some extent in the East Midlands, although the morphology of verbal endings in Midlands Middle English was rich enough to support the original Old English syntax. Further investigation will be needed to uncover why the CP-V2 grammar appears in the East Midlands. One possibility is that the collapse of agreement in that area, one of extensive Scandinavian settlement at the time of the 9th and 10th century Danish invasions, is subsequently reversed, due to contact with and population influx from adjacent dialect areas that maintained the native English morphology. This reversal could easily have happened without reversing the syntactic change from IP-V2 to CP-V2.
25. There is one major problem with saying that finite verbs in Mainland Scandinavian move to T: We would expect on such a proposal to find the verb moving across left-adjoined VP adverbs, but such movement seems never to occur. It is unclear how to interpret this fact, however, because it is hard to show that the languages allow left-adjunction to VP, and without such adjunction we cannot test whether the verb has moved out of VP to T. Evidence exists, in fact, that the Scandinavian languages resist the left-adjunction of adverbs to VP. Thus, in English certain aspectual adverbs, `completely', `entirely', and so forth, always occur VP-adjoined, either to the left or the right, and can never occur adjoined to higher projections. This is clear from the contrast between (ia) and (ib):
In Swedish, by contrast, the word order corresponding to (ia) is impossible, while that in (ib) can occur, though the sentence is less acceptable than one where the adverb is right-adjoined to VP. Here are illustrative examples from Anders Holmberg (personal communication to Bernhard Rohrbacher). The context is a non-CP-recursion subordinate clause to avoid interfering V2 effects on word order:
The contrast between English and Swedish is striking. We take it to show that left-adjunction to VP is blocked for some reason in Scandinavian, so that the adverb facts cannot be used to argue against movement to T.
26. If Mainland Scandinavian has verb movement to T and the negative particle is generated in a NegP above T, then we cannot maintain the widely-known analysis of object shift as movement to Spec,AGR-O. The problem is that movement of the direct object to Spec,AGR-O will always be string vacuous with respect to negation and adverbs adjoined above T. We assume that some other analysis will prove viable, perhaps one based on cliticization (see Bobaljik & Jonas 1993 and the references cited there). The clitic position would have to be higher than T in Scandinavian and lower than T in northern Middle English. Then when the verb moves to T in Scandinavian it will still be in a position that blocks object shift, while in Middle English it will have moved far enough to permit the object to move. This difference presumably reflects a general prohibition against overt material between TP and VP that Scandinavian seems to have, of which the prohibition against left-adjunction to VP discussed in note  is another manifestation.
27. The texts do, of course, exhibit phonological differences from West Saxon in their person endings, but no morphological differences. See Kroch et al. 1995 for further discussion.
28. The negated verbs in these examples are not relevant, as they would have moved to C-zero even in the southern dialect. The example from Luke is equivocal because the verb is interpretable as an imperative, though the Latin original has second person future.