Case # 35
Embedded Root Phenomena
1. Introduction and Background
2. What constitutes a root clause?
2.1. The canonical case: highest matrix clause
2.3. Other types of "parataxis"
4. Semantic/pragmatic explanations for the distribution of embedded root phenomena
5.1.1. The complementary distribution of overt complementisers and V2
5.1.2. Limited embedded V2: Frisian and Mainland Scandinavian
5.1.3. General embedded V2: Icelandic and Yiddish
5.1.4. Embedded V2 with obligatorily absent complementiser: German and Frisian
5.2.1. Generalized CP recursion vs. different "topic" positions
5.2.2. What limits/licenses CP recursion?
5.2.3. Some remaining questions
5.2.3.1. The status of the "so ... that" construction
5.2.3.2. The presence and absence of the overt complementiser
5.2.3.3. CP recursion and the position of the V2 constituent
1. Introduction and Background
Within modern syntactic theory, focus on the importance of a fundamental distinction between root and non-root clauses is primarily associated with Emonds 1969 .0 In this work, Emonds drew a crucial distinction between two types of transformations: structure-preserving and root transformations. The basic idea was that in the general case transformations could only move a node into a position in which a node of that category could be "base-generated" (an insight that has remained part of syntactic theory). In this sense, transformations are constrained to be "structure preserving". However, Emonds maintained that certain transformations did not obey this constraint; these transformations were however, limited in their application: they could only apply in a "root sentence", defined as "either the highest S in a tree, an S immediately dominated by the highest S or the reported S in direct discourse" (Emonds 1969: 6).
Three further developments are crucial to defining the question of "embedded root phenomena" as a current issue in syntactic theory. The first is the observation, first discussed in detail in Hooper & Thompson 1973, that Emonds was factually incorrect in his characterisation of the distribution of the phenomena he attributed to root transformations. As discussed below, these phenomena occur also in a subset of embedded contexts. One strand of the literature on embedded root phenomena - beginning with the Hooper & Thompson article itself - is an attempt to define this subset of embedded contexts, and to explain the principle underlying the definition (in particular, whether these contexts are to be defined in functional/pragmatic, semantic, or syntactic terms).
The second crucial development is the proposal, associated above all with the work of Hans den Besten (den Besten 1977, 1983), but also suggested to a greater or lesser extent in Higgins 1973, Williams 1974, Koster 1975, and Emonds 1976, that there is in fact no need for a stipulative distinction between root and non-root transformations. An important subset of Emonds' root transformations are analysed as involving movement of some element to the Comp[lementiser] position (or to the specifier of CP; the distinction between these two positions was not made in the early work); such movement is blocked by the presence of a lexical complementiser. In this perspective, the root/non-root distinction is only an artefact of the typical, but not entirely general correlation of lexical complementisers and embedded clauses. The particular phenomenon most famously reanalysed as a case of movement to Comp is the Germanic Verb Second (V2) phenomenon, in particular as evidenced in German and Dutch; the empirical strength of the analysis is the immediate explanation for the complementarity of V2 and the presence of a lexical complementiser.
The third advance in the research into embedded root phenomena was the determination that Hooper & Thompson's observations concerning the availability of root transformations in embedded clauses in English in fact extended in some cases at least to the other Germanic languages. In particular, in some of the Germanic languages V2 can cooccur with an overt complementiser, at least in a subset of contexts. Given the syntactic analysis of V2 just mentioned, the question of defining these contexts is now complicated by the question of the syntactic analysis of these cases of "embedded V2".
The structure of this case is as follows:
- Section 2 first sets out briefly the question of what constitutes a root clause.
- Section 3 gives a characterisation of some important "root phenomena" that have been discussed in the literature.
- Section 4 reviews proposals to explain the distribution of embedded root phenomena in semantic/pragmatic terms.
- Section 5 presents in more detail the phenomenon most prominent in the literature: that of embedded V2.
- Section 6 draws some general conclusions and suggests directions for future research.
2. What constitutes a root clause?
This section presents clear and less clear cases of "root" or non-embedded clauses:
- the most straightforward cases of the highest matrix clause;
- coordinations of such clauses;
- other "paratactic" constructions;
- non-restrictive relatives.
Then, in Section 3, some root phenomena are presented.
2.1. The canonical case: highest matrix clause
In the simplest case, a root clause is the highest clausal node in a tree (leaving vague precisely which extended projection is at issue). Importantly, given the proposals mentioned in Section 1 that the root/embedded distinction is typically an epiphenomenon, the crucial distinction being the presence or absence of a lexical complementiser, instances of root clauses introduced by a complementiser have been pointed out in the literature. (1a, b) are examples from Frisian (De Haan & Weerman 1986 : 98); (2) is an example from Dutch (den Besten 1983: 62), and (3) examples from Swedish (Andersson 1975: 53).
(1) a. Dat it nou altyd sa moast. that it now always so must 'Why does it always have to be this way?' b. Wat oft ik drinke woe. what whether I drink would 'What would I like to drink?'
(2) Gelachen dat we hebben! laughed that we have 'How we laughed!'
(3) a. Att du aldrig kan lära dig knipa käft. that you never can teach REFL keep quiet '[It's annoying that] you can never learn to keep quiet.' b. Om du bara kunde lära dig knipa käft. if you only could teach REFL keep quiet If you could only learn to keep quiet! 2.2. Coordination
Coordinately conjoined sentences that are not embedded are also typically taken to be root clauses, where at least "and", "or", and "but" and their translation equivalents are the canonical conjunctions:
(4) a. This book I read, and/but that one I didn't. b. Up it goes, or down he comes. Coordination is too large an issue to discuss here, although many aspects of it are clearly relevant. In particular, there is an extensive literature on various aspects of "asymmetric coordination" in the Germanic V2 languages, one class of which involves V2 (a "root phenomenon") occurring in the non-initial conjunct of an embedded coordination, as in (5a) from Höhle 1990: 222 (compare with the expected verb-final order in (5b)):
(5) a. Wenn jemand nach Hause kommt, und da steht der Gerichtsvollzieher vor der Tür... when someone to home comes and there stands the bailiff at the door 'When someone comes home, and the bailiff is standing there at the door...' b. Wenn jemand nach Hause kommt, und da der Gerichtsvollzieher vor der Tür steht... when someone to home comes and there the bailiff at the door stands 'When someone comes home, and the bailiff is standing there at the door...' Conversely, it has been stated that non-initial conjunct matrix clauses in Old English show some features of embedded clauses: in particular, that they are more likely to be verb/Infl-final (see e.g. Traugott 1972). However, Pintzuk (1991: 321-344) argues that second conjunct matrix clauses in Old English only show an increased rate of Infl-final order when the first conjunct is clearly Infl-final, and hence that the effect is due to parallelism.
2.3. Other types of "parataxis"
Coordination is the canonical case of "parataxis": a combination of clauses neither of which is subordinate to the other. There are however other cases that have been analysed in this way, even though they lack the symmetry that coordination (sometimes) displays. Three possible cases are adverbial clauses expressing causation, "so" exent clauses in English, and free dass clauses in German.
2.3.1. Adverbial clauses expressing causation
It has been observed that a number of languages have (at least) a pair of elements expressing a causal link, one of which is followed by a subordinate clause, the other of which is followed by what appears to be a root clause. Among others, this has been argued to be the case for parce que and car in French, because and for in English, weil and denn in German, därför att and ty in Swedish. The following discussion focusses particularly on German, because in this language the syntactic contrast between root and subordinate clauses is particlarly striking; as will be discussed in more detail in Section 5, root clauses are associated with V2 word order (the finite verb follows the first constituent in the clause), while in subordinate clauses the verb is final. While the manifestations of subordinate status are different in the other languages, however, the general pattern described below appears to be the same.
Most obviously a clause introduced by denn obligatorily shows V2 word order - the order characteristic of complementiserless root clauses.
(6) a. Er ist böse, denn er ist zu früh aufgestanden. he is angry for he is too early arisen 'He is angry, for he got up too early.' b. *Er ist böse, denn er zu früh aufgestanden ist he is angry for he too early arisen is 'He is angry, for he got up too early.' Further, a quantifier in the main clause can only very marginally bind a pronoun in the clause introduced by denn (7a); this is in contrast to an adverbial clause introduced by weil (because) (7b), but like the pattern found in clausal conjunction (7c):
(7) a. *Niemand(i) war böse, denn er(i) ist zu früh aufgestanden. No one was angry for he is too early arisen 'No one(i) was angry, for he(i) got up too early.' b. Niemand(i) war böse, weil er(i) zu früh aufgestanden ist No one was angry because he too early arisen is 'No one(i) was angry because he(i) got up too early.' c. *Niemand(i) war böse, und er(i) ist spät aufgestanden. No one was angry and he is late arisen 'No one(i) was angry, and he(i) got up late.' Another fact that distinguishes denn and its equivalents in other languages from a subordinator like weil and its equivalents is that a clause introduced by denn cannot appear in initial position:
(8) a. Julia war glücklich, denn alles war gut gegangen. Julia was happy for everything was well gone. 'Julia was happy, for everything had gone well.' b. *Denn alles war gut gegangen, war Julia glücklich. for everything was well gone was Julia happy 'For everything had gone well, Julia was happy.'
(9) a. Julia war glücklich, weil alles gut gegangen war. Julia was happy because everything well gone was. 'Julia was happy because everything had gone well.' b. Weil alles gut gegangen war, war Julia glücklich. because everything well gone gone was Julia happy 'Because everything had gone well, Julia was happy.' Elements like denn are sometimes classified as conjunctions on the basis of the above patterns. However, there are strong reasons for drawing a distinction between the two cases (essentially this point is made in de Haan 2001, and see also the comments on some English cases in Huddleston & Pullum 2002: 1319ff). The construction with denn shows some features which distinguish it both from coordination and from subordination. In particular, und (and) can not only coordinate constituents of different syntactic types, it can also coordinate subordinated clauses. In the following example, note both the V-final word order of both conjuncts and the binding of the pronouns by the quantifier keiner (no one).
(10) Keiner(i) hat gesagt, dass er(i) fertig sei, und dass er(i) nach Haus wolle. no one has said that he ready is(SBJ) and that he to home wants(SBJ) 'No one(i) said that he(i) was ready and that he(i) wanted to go home.' In contrast, it appears that denn clauses cannot be associated with subordinate clauses. Not only do clauses introduced by denn obligatorily show V2 order, as already discussed, but binding of a pronoun in the denn clause is never possible from outside the clause, suggesting that the denn clause is obligatorily attached at the highest level of the highest clause.
(11) *Keiner(i) hat gesagt, dass er(i) fertig sei, denn er(i) habe die Arbeit rechtzeitig angefangen no one has said that he ready is(SBJ) for he has(SBJ) the work on time begun 'No one(i) said that he(i) was ready, for he(i) had begun the work on time.' Note again the contrast with the behaviour of a clause introduced by weil (because):
(12) Keiner(i) hat gesagt, dass er(i) fertig sei, weil er(i) die Arbeit rechtzeitig angefangen habe. no one(i) has said that he(i) ready is(SBJ) because he(i) the work on time begun has(SBJ) 'No one(i) said that he(i) was ready because he(i) had begun the work on time.' Correlated with this is the fact that a denn clause, unlike for example an adverbial clause introduced by weil (because), cannot form part of the same phonological phrase (or "focus domain") as the clause to which it attaches, but constitutes an independent phonological phrase, as shown by the contrast between (13a) and (14a):
(13) a. Ich bin FROH, weil er gekommen ist. I am happy because he come is 'I am HAPpy because he came.' b. Ich bin FROH, weil er geKOMmen ist. I am happy because he come is 'I am HAPpy, because he CAME.'
(14) a. *Ich bin FROH, denn er ist gekommen I am happy for he is come 'I am HAPpy for he came.' b. Ich bin FROH, denn er ist geKOMmen. I am happy because he is come 'I am HAPpy, for he CAME.' Although here we see a contrast between weil and denn, it has been observed that at least in current speech, weil may also be followed by a clause with V2 order . Strikingly, when it does, it patterns just like denn: the clause it introduces cannot appear in initial position, pronouns it contains can only very marginally be bound from any other clause, and it constitutes an independent phonological phrase (Wegener 1993).
As can be seen from the glosses, the contrasts in between denn and weil clauses hold also for English clauses introduced by "for" and "because", modulo the lack of general V2 in English; similar facts in Swedish are observed in the corpus study of Teleman 1967, cited in Andersson 1975.1
Although this section focuses on adverbials expressing causation, similar phenomena have been observed with concessives; thus in spoken German V2 word order is also possible after obwohl (although) (Wegener 1993). For discussion and analysis of the different interpretations of these cases, and the interaction with phonological phrasing, see Verstraete 1998, 2002, and references therein.
2.3.2 "So" clauses in English
Parataxis is explicitly invoked in Hoeksema & Napoli 1993 to characterise the relation between the clauses in sentences like (15a) ("Para-So"), which they contrast crucially with the subordinative type in (15b) ("Sub-So").
Hoeksema & Napoli point out various ways in which the Para-So construction of (15a) behaves like coordination rather than subordination. Among others, these include the possibility of inversion in either or both clauses (contrast (16a,c) with (16b)) and the failure of an "any" quantified noun phrase in the first clause to bind a pronoun in the second (contrast (17a,c) with (17b)). 2
(16) a. Down fell Mary, so hot was the sun. b. *The sun was so hot that down fell Mary. c. Out went Mary and in came John.
(17) a. *Fred didn't hire anyone(i) that day, he was so fed up with him(i) b. Fred didn't need anyone(i) so badly that he would hire him(i) without an interview c. *I don't like anyone(i) and I certainly don't need him(i). As discussed below, locative inversion is a typical "root phenomenon", and so is expected in coordinated root clause, but not in a subordinate clause 3. The failure of binding can also be explained on the assumption that such binding requires c-command, and that nothing within the first conjunct of a coordination c-commands into the second.
On the other hand, the Para-so construction is argued to differ from ordinary coordination in a number of ways, including the fixed order of the two clauses 4, and the unacceptability of Across-The-Board movement. The latter point is illustrated in the contrast between (18a) and (18b):
(18) a. *Who(i) did you marry t(i), you loved t(i) so much? b. What(i) did John buy t(i) and Mary borrow t(i)? This construction therefore looks rather more like the constellation we have seen in the last section. As noted there, clauses with denn/for or weil+V2 order cannot precede their associated clause, and they also do not allow Across-The-Board movement
(19) *Who(i) did you marry t(i), for you loved t(i) so much? As will be discussed in Section 4.2.2., however, even the second clause in the "subordinative" version of this construction (e.g. 6b) shows some evidence of root phenomena.
2.3.3. Free dass clauses in German
One further case that might fall under the general rubric of parataxis is that of "free dass" clauses in German, as described in Reis 1997. This construction is illustrated in (20a,b); note however that in contrast to the denn clauses discussed above, these clauses are obligatorily verb-final.
(20) a. Ist denn etwas los, dass Max so schreit? is then something up that Max so shouts 'Is something up, that Max is shouting so much?' b. Er muss im Garten sein, dass er nicht aufmacht. he must in-the garden be that he not opens 'He must be in the garden, since he's not opening the door.' Although these clauses are introduced by the complementiser dass (that), Reis demonstrates that they do not behave like adverbial adjunct clauses. The following are only two of the various behaviors that she discusses. First, just as in the paratactic "so" construction, and denn/"for"-clauses, the order is fixed. Here the dass clause cannot occupy the clause-initial position.
(21) a. Du bist blöd, dass du kommst. you are stupid that you come 'You are stupid to come.' b. *Dass du kommst, bist du blöd. that you come are you stupid 'You are stupid to come.' Second, free dass clauses cannot appear as sentence fragments in question-answer pairs (20a). This is in contrast to clauses introduced by e.g. weil (22b), but similar to clauses introduced by denn (22c):
(22) Wieso/Warum ist Fritz blöd? Why is Fritz stupid? 'Why is Fritz stupid?' a. *Dass er Ernas Nerzmantel bezahlt that he Erna's mink coat pays for 'That he's paying for Erna's mink coat.' b. Weil er Ernas Nerzmantel bezahlt Because he Erna's mink coat pays for 'Because he's paying for Erna's mink coat.' c. *Denn er bezahlt Ernas Nerzmantel For he pays Ernas mink coat 'For he's paying for Erna's mink coat.' On the other hand, Reis characterizes these clauses as "relatively" rather than "absolutely" unintegrated clauses for three reasons. First, free dass clauses do not necessarily form a separate focus domain/phonological phrase; second, they occur before, rather than after, adverbial clauses introduced by weil (because); and third - perhaps most strikingly - they can contain a pronoun that is bound by a quantifier in the matrix clause, even in contexts that do not favour the extension of the binding domain of the quantifier to (the "telescoping" phenomenon (Fodor & Sag 1982, Roberts 1987, 1989, Groenendijk & Stokhof 1990, Poesio & Zucchi 1992)).
(23) a. Jeder(i) war blöd, dass er(i) darauf eingegangen ist. Everyone(i) was stupid that he(i) on it went in is 'Everyone was stupid to go along with it.'
b. Keiner(i) ist blöd, dass er(i) sich um die Zukunft seiner(i) Kinder Sorgen macht. no one(i) is stupid that he REFL about the future his(i) children worries makes No one(i) is stupid to worry about the future of his(i) children. 2.4. Non-restrictive relatives
One final environment whose root/non-root status is hard to determine is that of non-restrictive or appositive relatives. Various ways in which non-restrictive relatives behave like coordinated root clauses are discussed in Ross 1967 and subsequently in Emonds 1979; some of these are taken up in Fabb 1990. Again one might seek to appeal to the possibility of variable binding as a diagnostic for hierarchical position. Emonds pointed out the contrast between the availability of such binding in a restrictive, but not in a non-restrictive relative:
(24) a. *I gave every assistant(i), who loved his(i) uniform, a new one. b. I gave every assistant(i) who loved his(i) uniform a new one. However, Emonds' examples do not make the necessary point, since they instantiate a more general fact: QPs headed by "every" cannot be the antecedents of non-restrictive relatives at all, irrespective of binding:
(25) *I gave every assistant, who was working for very little pay, a new uniform. Fabb avoids this problem by looking instead at the distribution of the negative quantifier "any", which can be licensed by being c-commanded by "only" (Linebarger 1987). If "only" appears in the noun phrase, "any" can appear in a restrictive, but not a non-restrictive relative:
(26) a. *Only the tourists, who have any imagination, go to visit Sicily. b. Only the tourists who have any imagination go to visit Sicily. Fabb concludes from (26) that the non-restrictive relative cannot be an adjunct within the noun phrase, in contrast to the restrictive relative. However, the same data also make the further point that the non-restrictive relative cannot be in a position c-commanded by the noun phrase "only the tourists", given that (26a) contrasts also with (27)
(27) Only the tourists visited any sites of interest. This seems to rule out even a structure in which the restricted relative clause is adjoined to the full noun phrase. This conclusion is strengthened by cases of restrictive relatives that seem necessarily to be in this position, and in which "any" is licensed. 5
(28) Only [[a man] and [a woman]] who have had any experience of life together will be able to relate to this film. Further, a pronoun in a non-restrictive relative cannot be bound by a quantified noun phrase that clearly c-commands the modified noun phrase. 6
(29) Every student(i) sent the professor who had helped her(i) a card. *Every student(i) sent Professor Smith, who had helped her(i), a card. Fabb's own conclusion is that the relative clause is not in fact syntactically related to the surrounding clause at all, but is only interpreted with it at a level of discourse structure (a conclusion reached also in Sells 1985 and Haegeman 1988; see also Emonds 1979, McCawley 1982, Safir 1986 and Napoli 1989). It should however be noted that non-restrictive relatives in V2 languages do not show V2 (thus in German, for example, they are verb-final, as noted in Emonds 1979, p. 216).
3. Some root phenomena
Section 3.1 provides a selective summary of some of the structural root phenomena that have been discussed in the literature:
- Structural phenomena in English:
- VP preposing
- negative constituent preposing
- topicalisation
- left dislocation
- directional adverb preposing
- prepositional phrase substitution
- preposing around "be" and participle preposing
- subject auxiliary inversion
- direct quote preposing
- complement preposing
- right dislocation
- "if - then" conditionals
- Structural phenomena in other languages:
- verb-second in Germanic
- the distribution of the topic markers "wa" and "nun" in Japanese and Korean, respectively
Section 3.2 gives some examples of non-structural root phenomena: speaker-oriented adverbials and interjections.
Then, in Section 4, attempts to provide a semantic explanation for the distribution of these phenomena in embedded clauses are presented and discussed.
3.1. Structural phenomena
The phenomena discussed in Emonds 1969 and subsequently in Hooper & Thompson 1973 are primarily instances of movement to the left periphery of the clause. These include the following constructions:
(30) Mary plans for John to marry her, and marry her he will. Negative constituent preposing
(31) Never in my life have I seen such a crowd.
(32) This book you should read.
(33) This book, it has the recipe in it. Directional adverb preposing and prepositional phrase substitution (also referred to in the literature as locative inversion)
(34) a. Up the street trotted the dog. b. On the wall hangs a portrait of Mao. Preposing around "be" and participle preposing (also referred to as predicate fronting)
(35) a. More significant would be the development of a semantic theory. b. Standing next to me was the president of the company.
(36) Will James ever finish reading that book? Direct quote preposing 7
(37) "I won first prize," Bill exclaimed.
(38) Syntax and semantics are related, I think.
(39) You should go to see it, that movie. Emonds 1969 and Hooper & Thompson 1973 also discuss some other phenomena, for which the reader is referred to the original works.
More recently, Iatridou 1991 has argued that "if - then" conditionals (in distinction to conditionals where the consequent is not introduced by "then") are also a root phenomenon. In her analysis, "then" occupies the Comp position in the consequent clause, and the "if" clause the Spec[CP] position. Thus, although no movement is involved, under most analyses this case is predicted to fall together with e.g. left dislocation or Negative Inversion, since the Spec[CP] position is necessarily filled.
The best studied "root phenomenon" is not however from English at all, but is the V2 construction found in the other Germanic languages. In languages such as German, Dutch, and Frisian the asymmetry between the position of the verb in main and subordinate clause is particularly striking, as illustrated by the German examples in (40)
(40) a. Er weiß nichts davon. he knows nothing of it 'He knows nothing about it.' b. ...dass er nichts davon weiß. ...that he nothing of it knows '...that he knows nothing about it.' Since den Besten 1977, 1983, the most widely accepted account of this contrast is that German and Dutch are OV and Infl-final but Comp-initial, and that when the complementiser position is not filled by an overt element, the finite verb moves to occupy it; further, some other constituent moves to Spec[CP]. For much more detail and alternative analyses, <link to case on V2>
The mainland Scandinavian languages (Danish, Norwegian, and Swedish) show a less dramatic asymmetry, since they are not only Comp-initial, but also VO and Infl-medial (though it is not clear whether the verb moves out of the VP in finite clauses unless it continues to move to Comp). However, given these differences, the same assumptions about movement of the finite verb to Comp and of some other element to Spec[CP] can account elegantly for the difference in the order of the negative element and the verb in matrix and subordinate clauses ((41a,b) and (42a,b)), and also for the restriction of non-subject-initial clauses to root position (see in particular Vikner 1995 for analysis and a detailed description of the data). These examples are from Swedish:
(41) a. Lena köpte inte en ny bok igår. Lena bought not a new book yesterday 'Lena didn't buy a new book yesterday.' b. *Lena inte köpte en ny bok igår. Lena not bought a new book yesterday 'Lena didn't buy a new book yesterday.'
(42) a. *Han beklagade att Lena köpte inte en ny bok igår. he complained that Lena bought not a new book yesterday 'He complained that Lena didn't buy a new book yesterday.' b. Han beklagade att Lena inte köpte en ny bok igår. he complained that Lena not bought a new book yesterday 'He complained that Lena didn't buy a new book yesterday.'
(43) a. Dom betvivlar att vi tycker IFK är bäst. They doubt thatt we think IFK is best 'They doubt that we think IFK is best.' b. ?Dom betvivlar att IFK tycker vi är bäst. They doubt that IFK think we is best 'They doubt that we think IFK is best.' English allows only very limited V2; the most uncontroversial cases are negative preposing and matrix questions with subject-aux inversion. In Standard English both are root phenomena; subject-aux inversion in embedded questions is however found in a number of Hiberno-English dialects: see Henry 1995 for Belfast English, and McCloskey 1992 for other varieties.
Finally, the analysis of V2 in Germanic has inspired similar analyses in other unrelated languages. It has been observed that in Japanese and Korean, topic phrases marked with "wa" and "nun" respectively, and interpreted non-contrastively, are disallowed in most embedded contexts, and are thus often taken to be root phenomena. Whitman 1989 argues that the distribution of these topic markers is paralleled exactly by the distribution of clause-final modal particles, although this correlation is not completely uncontroversial. He uses this fact to argue that in these languages topics appear in Spec[MoodP]. Whitman further notes that these non-contrastive topics do in fact occur in a number of embedded contexts, and argues for an analysis along the lines of that proposed for Yiddish in Diesing 1990 (see Sections 5.1.3, 5.2.1). His data and analysis are however disputed in Choi 2000.
3.2. Non-structural phenomena
To Emond's (1969) and Hooper & Thompson's (1973) list of root phenomena involving movement, Green 1976 adds some phenomena (again, only from English) which do not involve movement but rather the felicity of certain adverbial adjuncts and interjections, typically having to do with speaker attitude. The following are some examples; again the reader should refer to the original article for additional cases not listed here.
(44) [Assuming languages can have nasal assimilation rules, this is a natural analysis.]
Indeed, languages must have nasal assimilation rulesFlat "indeed"
(45) [Do you have any children?]
Indeed I do."Lo and behold"
(46) Lo and behold, there was a unicorn among the roses. "Frankly"
(47) Frankly, Bobby Riggs never had a chance. As Green herself points out, the degree to which these expressions are really restricted to root clauses varies.
Further cases are given in Banfield 1982, and similar examples can be found for other languages. Reis 1997 gives examples with German nämlich (in fact), and Andersson 1975 discusses various adverbials and interjections in Swedish including för guds skull (for God's sake), din idiot (you idiot), tyvärr (unfortunately), and härmed (hereby).
4. Semantic/pragmatic explanations for the distribution of embedded root phenomena
As mentioned in the introduction, the issue that arises as soon as "embedded root phenomena" are recognized is whether there is any systematicity to their distribution, and if so, how it can be characterised. In their work on embedded root phenomena, Hooper & Thompson propose a semantic characterisation of the distribution of root phenomena: these occur only in clauses that are asserted (Hooper & Thompson 1973: 472). By their own admission, Hooper & Thompson do not give "an absolute definition" of what constitutes an asserted clause. They state that the assertion of a sentence is "its core meaning or main proposition," and that it "may be identified as that part which can be negated or questioned by the usual application of the processes of negation and interrogation". Sentences may contain more than one assertion (in the case of a coordination, for example); also (crucially), some subordinate clauses are asserted. Based on this notion of assertion, they provide a 5-way division of predicates taking sentential complements or sentential subjects:
- Class A predicates (e.g. "say", "report", "be true", "be obvious"). The verbs in this group are all verbs of saying. Both the verbs and the adjectives in this group can function "parenthetically", in which case the subordinate clause constitutes the main assertion of the sentence. It is claimed however that if the subordinate clause occurs in subject position (as in, e.g. That German beer is better than American beer is true) it is not asserted.
- Class B predicates (e.g. "suppose", "expect", "it seems", "it appears"). In this group also the predicates can function parenthetically, and in this case the subordinate clause is asserted. The distinction between this group and Group A is not made entirely clear, although it is noted that Class B predicates allow "Neg raising" and tag questions based on the subordinate clause.
- Class C predicates (e.g. "be (un)likely", "be (im)possible", "doubt", "deny") have complements which are not asserted.
- Class D predicates (e.g. "resent", "regret", "be odd", "be strange"); these factive predicates have complements which are argued to be presupposed, and hence not asserted.
- Class E predicates (e.g. "realize", "know"); these semifactives (factives that lose their factivity in questions and conditionals) have a reading on which the subordinate clause is asserted.
Further, Hooper & Thompson claim that "reduced" clauses (infinitives, gerunds, and subjunctive clauses) are never asserted, nor are noun complement clauses. Restrictive relative clauses on definite heads are argued to be presupposed (and therefore never asserted); non-restrictive relatives, and restrictive relatives on indefinite heads, are not presupposed (and may be asserted). Adverbial clauses may or may not be asserted (and in some cases may force the main clause itself to be read as presupposed, and hence not asserted (pp. 486-495)).
In addition to claiming that the environments allowing root phenomena must be characterised in terms of a semantic concept of "assertion", Hooper & Thompson claim that there is a pragmatic explanation for the restriction of "root transformations" to these environments: these transformations, it is claimed, all have the function of emphasising a particular constituent (p. 470), and "emphasis would be unacceptable in clauses that are not asserted" (p. 472).
Hooper & Thompson's claim that root phenomena can occur in all and only asserted clauses is disputed in Green 1976. Green points out that in at least some cases their claim that a particular clause is asserted is backed up only by the grammaticality of some root transformation occurring in it - clearly a circular argument- and also that some root phenomena occur in the complements to the counterfactive "pretend" and to the performatives like "bet", "promise", and "predict" (pp. 390-91). Green herself puts forward a pragmatic hypothesis - that embedded root phenomena are licensed "just in case the proposition they affect, and therefore emphasize, is one which the speaker supports" (p. 386), but she argues that this is only one constraint out of many affecting the acceptability of these phenomena.
It is a general problem for work in this area that definitions given are vague and independent evidence for the validity of the concepts used often weak. Subsequent researchers have proposed various modifications of Hooper & Thompson's proposal that "assertion" is a necessary and sufficient condition for the occurrence of root phenomena. Based on data from Swedish, Andersson 1975 argues that the relevant semantic distinction is between clauses that make a statement, ask a question, or give a command (his class of "semantically main clauses") and those that perform none of these functions ("semantically subordinate clauses"). He also proposes that the non-negative predicates in Class C belong more properly in Class B, while negated Class B predicates belong in Class C. Later work on Germanic has sometimes included further variants of these proposals (see for example Iatridou & Kroch 1992), although typically such work is restricted to finding a basis for the classification of predicates that allow root phenomena in their complements, rather than finding a single semantic concept that will also extend to other syntactic environments ("reduced" clauses, relatives, clauses in subject position, etc). This is because work on V2 in particular (including that of Andersson 1975) has led to the conclusion that there is an irreducibly syntactic aspect to the distribution of at least the structural root phenomena. A detailed discussion of the verbs in Danish and German that do and do not allow V2 in their complements can be found in Vikner 1995: 71f (and see references therein); Vikner's own conclusion is that there is no single semantic concept that can predict the patterns he reports.
The second aspect of Hooper & Thompson's analysis - that the failure of root transformations to occur in non-asserted clauses is due to the pragmatic incompatibility of emphasis with non-assertion - has not been pursued in subsequent literature. In fact there is counterevidence to this claim in their own article. They point out that emphasis can be achieved by other means, including clefting (It's this book that you should read), and they further point out that clefting is a "structure preserving transformation," so that Emonds' 1969 proposal would predict, correctly, that it can occur quite generally in embedded contexts. But their proposal appears to be precisely an attempt to to derive a (modified version of) Emonds' generalisation about the limited distribution of "non-structure preserving" transformations from an interaction of semantic and pragmatic considerations:
[...W]e can say that these transformations operate only on Ss that are asserted. R[oot] T[ransformation]s are not applicable in presupposed sentences because it is not appropriate to emphasize elements of a sentence whose proposition is already known, whose truth is presupposed, and whose content is relegated to the background.
[... A] definition of root transformation in terms of its linguistic function rather than in terms of syntactic structures enables us to explain the otherwise mysterious facts about its applicability. (p. 496)
If the explanation for the distribution of root transformations is really entirely semantic/functional, then it should apply equally to other constructions that have the same semantic/functional properties, regardless of their status as "structure preserving". 8 Hence the felicity of e.g. clefting in presupposed clauses (e.g. We regretted that it was precisely this book that had been destroyed) does constitute evidence against this aspect of their analysis.
5. Embedded Verb Second
This section discusses in some detail the most-studied syntactic root phenomenon: Germanic V2.
- Section 5.1 sets out the main characteristics of embedded V2
- Section 5.2 presents and discusses analyses that have been proposed
Then, in Section 6, some general conclusions are drawn, and directions for further research suggested.
5.1. The phenomena
This section discusses four main aspects of embedded V2:
- the original observation that the correct generalisation about V2 is that it is in complementary distribution with complementisers;
- languages in which this complementarity does not hold in a restricted set of embedded clauses (Frisian, Mainland Scandinavian);
- languages in which this complementarity has been argued not to hold in general (Icelandic, Yiddish);
- the distribution of embedded V2 clauses in German.
5.1.1. The complementary distribution of overt complementisers and V2
The V2 phenomenon itself is sketched above in Section 3.1, and discussed in much greater detail in Case V2 <link to V2>. As mentioned in Section 3.1, following the work of den Besten, it has been generally accepted that the restriction of V2 to root sentences is in fact an epiphenomenon; the real asymmetry is between sentences with and without overt complementisers. V2 is analysed as the effect of the tensed verb moving to the empty Comp position, and some XP moving to Spec[CP].
This analysis predicts straightforwardly the complementary distribution of an overt complementiser and verb-fronting found in German. However, there are a set of exceptions to this complementarity in a number of other Germanic languages, which have been grouped into two classes. Mainland Scandinavian and Frisian fall into one class, Yiddish and Icelandic into another.
5.1.2. Limited embedded V2: Frisian and Mainland Scandinavian
This section sets out the distribution of embedded V2 in Frisian and Mainland Scandinavian to the extent that it has been established to be similar or identical in these languages. Some possible points of difference will be discussed in Section 5_2_3_3. For reasons that will become clear, this pattern is referred to by Vikner (1991, 1995) as "limited embedded V2".
In de Haan & Weerman 1986 it is pointed out that in Frisian, a verb-final West Germanic language, V2 can occur in subordinate clauses introduced by a complementiser:
(48) ik leau dat hy kin him wol rêde. I believe that he can himself surely save 'I believe that he can take care of himself.' However, the occurrence of such clauses is restricted.
- First, they occur after verbs of saying and thinking, such as sizze (say), roppe (call), tinke (think), leauwe (believe), but not after hoopje (hope), wite (know), ferwachtsje (expect), or spite (regret).
- Further, they cannot follow even verbs in the first group if these are negated, modalized (de Haan & Weerman 1986 ) or questioned (de Haan 2001).
- They are not in general possible as adjuncts (with the exception of rationale clauses introduced by om't, omdat, or omreden dat, and extent clauses licensed by sa+degree phrase (de Haan 2001: 6))
- They cannot occur in subject position (Iatridou & Kroch 1992).
- Nor can they be moved into the initial "topic" position from complement position (de Haan 2001: 10).
The last point is illustrated in (49); this example also illustrates a root phenomenon particular to Frisian: the subject pronoun in the subordinate clause is different in the two sentences as the clitic er cannot appear in the position immediately following the complementiser in an embedded V2 clause, just as it cannot appear in the initial position of a root clause.
(49) a. Dat er it antwurd net witen hie, sei er. that he the answer not known had said he 'He said that he hadn't known the answer.' b. *Dat hy hie it antwurd net witten, sei er. that he had the answer not known said he 'He said that he hadn't known the answer.' In Mainland Scandinavian (Danish, Norwegian, and Swedish) the distribution of embedded V2 is very similar to that of Frisian. V2 clauses introduced by a complementiser can only appear as the complements of a restricted set of verbs (as in Frisian, these verbs are a subset of the "bridge" verbs that allow extraction). This restriction is discussed in detail for Danish in Vikner 1991 , 1995, where it is noted however that the exact list of verbs that allow embedded V2 is not identical from language to language (Vikner 1995: 70-72). In particular, as in Frisian, inherently negative verbs (such as doubt deny, regret) or verbs taking irrealis complements do not license embedded V2; and again, even the subset of bridge verbs that allow embedded V2 do not license it when negated or modalized. Whether or not questioning the verb in the matrix clause makes V2 impossible in its complement clause is not clear from the available literature. As in Frisian, V2 clauses are not possible as adjuncts, as sentential subjects (Iatridou & Kroch 1992), or as topicalised complements. Good exemplification of the data for Swedish can be found in Andersson 1975.
5.1.3. General embedded V2: Icelandic and Yiddish
In Icelandic and Yiddish it has been argued that V2 is found generally in subordinate clauses. There is no restriction to the complement position of a particular set of "bridge verbs"; V2 clauses can also appear in a variety of non-complement clauses, including relative clauses, and clauses in adjunct and subject position (Diesing 1998, 1990, Rögnvaldsson & Thráinsson 1990, Vikner 1991, 1995, Santorini 1989 , 1992, Iatridou & Kroch 1992).
5.1.4. Embedded V2 with obligatorily absent complementiser: German and Frisian
The instances of embedded V2 mentioned in Sections 5.1.2 and 5.1.3 are a challenge to the analysis of V2 as involving movement of the finite verb to an unfilled complementiser position, since in all the cases discussed there V2 takes place below an overt complementiser. German also allows embedded V2, but in this case it is only possible when the complementiser is absent, as predicted by the standard analysis outlined in Sections 3.1 and 5.5.1 (Vikner 1995 ). In fact in Frisian also embedded V2 clauses are only optionally introduced by the complementiser dat (that); further, there are dialects of Frisian in which, like German, embedded V2 is only possible if the complementiser is absent (de Haan 2001, Sipma 1913). The following data are from German:
(50) a. Sie sagte, daß sie keine Bücher kaufen wolle. she said that she no books buy wants(SBJ) 'She said that she didn't want to buy any books.' b. Sie sagte, sie wolle keine Bücher kaufen. she said she wants(SBJ) no books buy 'She said she didn't want to buy any books.' c. *Sie sagte, daß sie wolle keine Bücher kaufen. she said that she wants(SBJ) no books buy 'She said that she didn't want to buy any books.' d. *Sie sagte, sie keine Bücher kaufen wolle. she said she no book buy wants 'She said she didn't want to buy any books.' In other respects, German is typically described as being more like the "limited V2" languages discussed in Section 5.1.2: embedded V2 only occurs in the complements of a subset of "bridge" verbs (Vikner 1995, Reis 1997), and subordinate clauses with V2 order cannot appear in subject or topic position (Reis 1997).
There are however a number of exceptions to these generalisations (Beatrice Santorini, personal communication). Some nouns allow V2 complements:
(51) der Glaube, die Erde sei flach the belief the earth is(SBJ) flat 'the belief that the earth is flat' V2 clauses seem also to appear in topic position:
(52) Er habe nichts zu verlieren, sagte er mir he has(SBJ) nothing to lose said he me(DAT) 'He had nothing to lose, he told me.' However, when a V2 clause does appear in the initial position as in (53b), it may not contain a pronoun bound by a quantifier in the other clause (Reis 1997: 139), unlike a V2 clause that has been extraposed to the right (53a) or a verb-final clause whether extraposed or in initial position (53c):
Reis's conclusion is that examples like (52) and (53b) are not topicalizations of complement clauses, but that the V2 clauses are main clauses, with the verb of saying part of a V1 parenthetical (note that the subjunctive may appear in root clauses in German).
It should also be noted that in German V2 clauses may appear as the complements to verbs that are negated (54a), modalised (54b), and questioned (54c):
(54) a. Er sagte nicht, er sei fertig. he said not he is(SBJ) ready 'He didn't say he was ready.' b. Er würde sagen er sei fertig. he would say he is(SBJ) ready 'He would say he was ready.' c. Sagte er, er sei fertig? said he he is(SBJ) ready 'Did he say he was ready?' The similarity in distribution of German embedded V2 and the limited embedded V2 found in Frisian and Mainland Scandinavian described in Section 5.1.2, is therefore only partial.
The description of embedded V2 in Frisian that has been given so far is in fact an oversimplification: in addition to the pattern described in Section 5.1.2, Frisian also displays the German pattern. That is, if the complementiser is absent, and if there is no intonation break, embedded V2 is possible even if the matrix verb is negated or modalised, and binding of a pronoun within the complement clause is also grammatical (de Haan 2001: 34).
Overall, the German pattern of embedded V2 appear to provide strong evidence for the idea that V2 is not a "root" phenomenon, but derives instead from the absence of an overt complementiser, an absence which does not correlate perfectly with "root" status. Rather surprisingly, in den Besten 1983 this conclusion is not drawn; instead den Besten claims that the type of sentence illustrated in (50b), (54a-c) is not in fact a subordinate clause at all, but rather that it has essentially the syntax of a direct quotation, despite the occurrence of the subjunctive and the pronominalisation pattern. There are however strong arguments against this conclusion, including the fact that some of the verbs that can be followed by V2 complements do not allow direct quotations (e.g. annehmen (realize) (Reis 1997). Thus, alongside the examples in Section 2.1 of "non-root" phenomena occurring in "root" clauses, just when these have an overt complementiser (or a wh-word), these German cases appear to be perfect examples of a root phenomenon occurring in embedded clauses just in case these lack an overt complementiser (or wh-word).
5.2. Analyses and issues
This section presents and discusses some of the important issues that have arisen in the analysis of the phenomena discussed in the last section.
- Does V2 under a complementiser always involves recursion of CP, or may it result from a topic position lower than [Spec,CP]?
- What licenses CP recursion?
- Three remaining theoretical questions:
- How can the behaviour of the "so ... that" construction be reconciled with generalisations concerning the distribution of CP recursion?
- How can an account of embedded V2 in terms of the licensing of CP recursion simultaneously explain the obligatory absence of a lexical complementiser in German and its obligatory presence in Mainland Scandinavian?
- What is the attachment site of the V2 clause?
5.2.1. Generalized CP recursion vs. different "topic" positions
There is widespread agreement in the literature that the type of embedded V2 found in Mainland Scandinavian and Frisian (limited embedded V2 below an overt complementiser) is to be explained by the hypothesis that there can be recursion of CP: that is, that the overt complementiser takes as its complement another CP. This clearly allows for the retention of the idea that the finite verb in a V2 clause is in C (in this case, the head of the lower CP projection) and the initial XP in the Specifier of this lower CP. This hypothesis is defended in Holmberg 1986 , Platzack 1986 for Swedish; de Haan & Weerman 1986, Iatridou & Kroch for Frisian; Vikner 1991 , 1995, Iatridou & Kroch 1992 for Danish.
More controversial is the correct analysis of the type of embedded V2 found in Icelandic and Yiddish (general embedded V2). Various analyses have been put forward, but they can be divided into two types of approach. In the approach defended most extensively by Vikner (1991, 1995), general and limited V2 should be given a unitary analysis: both types involve CP recursion. In all cases the finite verb is in C and the XP preceding it in the Specifier of this CP; the complementiser is the head of a higher CP. Vikner argues that general embedded V2 arises in a V2 language if and only if the language has both a right-branching IP (e.g. all Scandinavian languages and Yiddish, but not for example, Dutch, Frisian, or German), and movement of the finite verb to I (e.g. Icelandic, Yiddish, but not Mainland Scandinavian). The reason for this correlation, Vikner proposes, is that a child exposed to such a language has no basis for postulating a difference between root and non-root clauses (Vikner 1995: 159-163).
In the second approach to general embedded V2, it is assumed that in these languages the domain of V2 is not CP, but some smaller clausal constituent. Thus Diesing 1988 , 1990, Santorini 1989, 1992 have argued for Yiddish, and Rönvaldsson & Thráinsson 1990 for Icelandic, that in these languages the finite verb in a V2 sentence is in I, and the clause-initial XP in Spec[IP], the subject remaining in its VP-internal subject position. Given the explosion of I into a number of distinct functional heads, there are variants of this analysis according to which the subject does not remain within VP but moves into a higher specifier, and the finite verb occupies one of these functional head positions (see for example Thráinsson 1994); the crucial point here, however, is that the verb does not occupy the Comp position, so that no complementary distribution of V2 and overt complementiser is expected.
5.2.2. What limits/licenses CP recursion?
Although there has been considerable debate as to whether CP-recursion is the correct analysis for Yiddish and Icelandic, there has been much less discussion in the literature of 80s and 90s of the basis for the limitations on CP-recursion found in those languages for which it has become a relatively standard analysis for embedded V2 - that is, the languages with limited embedded V2. If CP is a possible complement for Comp, why are there language-internal restrictions on this recursion? Given the demonstrated similarity in these restrictions, what underlying principle could be responsible? Essentially this is a revised version of part of the question addressed in Hooper & Thompson 1973: what is the correct characterisation of the contexts licensing CP-recursion, and does this characterisation lead to an explanation?
Both Authier 1992 and Iatridou 1991 point out that CP recursion in English declarative complements is only possible in environments that also allow "that" deletion (i.e. environments where a declarative subordinate clause need not be introduced by an overt complementiser). Given prior arguments that "that" deletion is only possible in CPs governed by a verb, they conclude that this is also a necessary condition for CP recursion. However, more must be said to explain the absence of CP recursion in e.g. the complements to negative verbs. Basing themselves on the proposal in Laka 1990 that negative and negated verbs select a negative complementiser (and Hegarty's related 1992 proposal for certain factive verbs), Iatridou & Kroch 1992 propose that only semantically empty CPs - or at least, those whose content can be recovered from the lower CP - can be deleted at LF, and only when governed by a verb. If there are two CPs and the higher cannot delete (because there is material in its specifier, or because it has content that cannot be so recovered) then the lower CP is unlicensed.
5.2.3. Some remaining questions
5.2.3.1. The status of the "so...that" construction
One known exception to the generalisation that CP recursion is only possible in positions governed by a verb is a particular type of extent clause in Frisian (the following example is from de Haan & Weerman 1986):
(55) Hy is sa meager (dat) hy kin wol efter in reid skûlje. he is so thin (that) he can well behind a cane hide 'He is so thin that he can hide behind a cane.' Such examples clearly do not fall under Iatridou & Kroch's (and Authier's) generalisation that CP recursion is possible only under government from a verb. Observing that this construction does not license CP recursion in Danish, Iatridou & Kroch conjecture that dat (that) in Frisian may have a second lexical entry as a "coordinating particle". However, it should be noted that the same construction in English also shows some signs of allowing CP recursion (contra the claims of Hoeksema & Napoli 1993, discussed in Section 2.3.2, that this is a clear case of subordination):
(56) a. This kind of equation is so complex that if it has one solution, then it probably has several ("if-then") b. She was so happy that off she ran without looking behind her. (locative inversion) c. Humphrey was so rich that in no circumstances would he use anything other than cash. (negative preposing) When there is no evidence for CP recursion this construction does not require a complementiser in English, so at least that correlation is maintained (the putative restriction of complementiserless subordinate clauses in English to the complements of bridge verbs is however known to admit relatively common counterexamples, and intuitions about e.g. complement clauses to nouns are not strong). Thus the example in note 4, repeated here as (57) could potentially either be taken to be an "inverted" case of the paratactic "so" construction or an instance of subordination plus "complementiser drop":
(57) The sun was so hot she fainted. Dutch also allows V2 in the extent clause when the complementiser is absent (Jack Hoeksema, personal communication): 9
Van Kemenade (1997: 342) also cites in passing an example of embedded verb movement to C in early Middle English in an extent clause of this type:
(59) þat al is hare blisse se muchel þat ne mei hit munne na muð that all is their bliss so great that not may it mention no mouth 'that all their bliss is so great that no mouth may mention it' German, on the other hand, does not allow V2 in an extraposed extent clause (Beatrice Santorini, personal communication). Thus (60b) contrasts with its Frisian and Dutch counterparts in (55) and (58b) above.
However, the German counterpart of (58a), where the extent clause is initial, is fully grammatical:
(61) Er kann sich hinter einem Schilfrohr verstecken, so mager ist er. He can REFL behind a cane hide so thin is he 'He can hide behind a cane, he is so thin.' This construction then poses unresolved problems for attempts to provide a unified analysis of the distribution of CP recursion.
5.2.3.2. The presence and absence of the overt complementiser
As discussed in Section 5.1.4, embedded V2 in German occurs only in the absence of a complementiser. Vikner (1995: 84) points out that there is therefore no reason to hypothesize any kind of CP-recursion in this case. However, there is at least some degree of similarity in the distribution of embedded V2 in German and that found in the other "limited embedded V2 languages" (Section 5.1.2), although as noted in Section 5.1.4 there are limits to the similarity. If we assume that the similarity needs to be accounted for, we have to ask how this is to be achieved. The proposal in Iatridou & Kroch 1992 suggests a possible analysis. As just discussed, this account adopts Laka's hypothesis of a distinct negative complementiser (and in addition also irrealis and factive complementisers); in contrast to Basque, however, where there is an overt morphological distinction, these complementisers in e.g. Mainland Scandinavian must be homophonous with the "unmarked" affirmative declarative one. If we assume that in German the phonetically empty Comp head that hosts the features inducing V2 also carries these affirmative declarative features, but that only the overt complementiser instantiates the negative, irrealis, or factive features, this could explain the lexical restriction on embedded V2.
What would still remain unexplained under this proposal is the unacceptability of embedded V2 in German in positions not governed by a verb. In Iatridou's and Kroch's account, the lack of CP recursion in e.g. sentential subject position, or the Spec[CP] topic position, follows from the hypothesis that only government by a verb allows for the LF deletion of the topmost CP. But this proposal would not extend to the German case, if V2 in this language does not involve CP recursion at all. An alternative would be to assume instead that the affirmative declarative features in the lower Comp (in the CP recursion context) or on the null Comp (in the German case) have to be licensed under government by a verb.
The issue just raised for German was how to explain the similar distribution of embedded V2 in that language despite any evidence for a recursive CP. Conversely, it also remains to be explained why in the mainland Scandinavian languages and English (but not Frisian: see Section 5.1.4) root phenomena in embedded CPs require the presence of an overt complementiser (see Grimshaw 1997 for an explanation for the English case, and Vikner 1995: 84f for a brief discussion of the Danish facts illustrated in (63), where the complementiser is obligatory only if V2 has applied; the situation is different in Swedish: the complementiser
att may be omitted whether or not the following clause shows V2.)
(62) a. She believed (that) he would never let her down. b. She believed *(that) never would he let her down.
(63) a. Hun sagde ??(at) vi skulle ikke købe denne bog. she said (that) we should not buy this book 'She said that we shouldn't buy this book.' b. Hun sagde (at) vi ikke skulle købe denne bog. she said (that) we not should buy this book 'She said that we shouldn't buy this book.' There are two aspects to this question: the first is why CP recursion is required (that is, why do these languages not behave like German (Section 5.1.4)). Even if this is explained, however, a second question arises: why should the higher C delete only at LF and not at PF also, given that some kind of PF Comp deletion has been taken to be the explanation for the optional appearance of "that" in sentences like (62a) above. A possible explanation for English might be that PF Comp deletion is not the right analysis for (62a), but rather that we need to assume the existence of a zero complementiser. As proposed in Rizzi 1990, for quite different reasons, this complementiser has a rather richer feature content than its apparent overt counterpart "that", since it can agree with its specifier (this is the basis for his account of the "that-trace" phenomenon). If this richer feature content were enough to prevent LF-deletion, that would explain its failure to license CP recursion.
5.2.3.3. The position of the V2 constituent
Leaving aside Yiddish and Icelandic (the languages in which V2 appears in all types of subordinate clause, and which therefore does not appear to be a root phenomeonon), we have found the following examples of embedded V2:
- V2 in certain adverbial clauses (most commonly those expressing causation)
- V2 after an optional complementiser (dat) in extent clauses in Frisian and possibly some other languages
- V2 after an (obligatory) (Danish, Norwegian(?)), or optional (Frisian, Swedish) complementiser in the complement of certain verbs
- V2 in the complement of certain verbs only when the complementiser is absent (German, Frisian)
We have seen that there is evidence that the type in (1) is attached at a very high point in the structure of the matrix clause (in fact all the evidence indicates that it must be high enough to be outside the scope of any element in the root clause). The same is true for the type in (2).
WIth respect to the type in (4) Reis 1997 argues for German (and de Haan 2001 duplicates most of the arguments for the corresponding cases in Frisian) that the attachment site is relatively low: in particular, such clauses obligatorily precede adverbial clauses. This is true even when the latter take scope lower than the subject. In (64), in addition to a reading where the rationale clause modifies the V2 complement clause, there is a grammatical reading where it modifies the matrix - but must still be attached low enough to allow the pronouns in it to be bound.
(64) Keiner(i) hat gesagt, er(i) komme, bloß weil er(i) seine(i) Mutter glücklich machen wollte. no one has said he comes just because he his mother happy make wanted 'No one said he was coming just because he wanted to make his mother happy.' Further, as (64) shows and as Reis also points out, a quantifier in the matrix can bind a pronoun in the V2 clause, again suggesting that the V2 clause attaches low (although it cannot topicalise with the VP, unlike a complement (p. 140)). Reis's own conclusion is that embedded V2 clauses of this type have exactly the same status (and occupy the same hierarchical position) as the free dass clauses discussed above in Section 2.3.3.
What about the cases in (3): that is, embedded V2 in complement clauses in the Mainland Scandinavian languages and the subset of cases in Frisian where the complementiser can appear? De Haan 2001 argues that the Frisian cases should be assimilated to the type in (1) and (2), and that in all cases the position of the "embedded" V2 clauses is very high. In fact, he argues that the V2 clause has to attach at the same level as a root CP, to which it stands in a paratactic relation. Among other evidence, in Frisian a quantified subject in the matrix cannot bind a pronoun in the V2 clause, as shown by the contrast in (65):
(65) a. [Ider fan ús](i) sei dat er(i) it net wist. each of us said that he it not knew '[Each of us](i) said that he(i) didn't know it.' b. *[Ider fan ús](i) sei dat hy(i) wist it net. each of us said that he knew it not '[Each of us](i) said that he(i) didn't know it.' Further, the clause containing the verb selecting the recursive CP must itself be a V2 clause:
(66) a. Niis sei Teake (dat) hy hie it net witten. a minute ago said Teake (that) he had it not known 'A minute ago Teake said that he hadn't know that.' b. *Ik tocht dat Teake niis sei (dat) hy hie it net witten. I thought that Teake a minute ago said (that) he had it not known 'I thought that Teake said a minute ago that he hadn't known that.' Preliminary investigation suggests that Dutch embedded V2 is similar to Frisian in these respects. A matrix quantifier cannot bind a pronoun in a V2 extent clause in the "so... that" construction (67b), or in a V2 complement to a verb of assertion (68b) (Jack Hoeksema, personal communication). 10
(67) a. Iedereen(i)/niemand(i) was zo blij, dat hij(i) een gat in de lucht sprong. everyone/noone was so happy that he a hole in the sky jumped 'Everyone(i)/Noone(i) was so happy that he(i) jumped for joy.' b. *Iedereen(i)/niemand(i) was zo blij, hij(i) sprong een gat in de lucht. everyone/noone was so happy he jumped a hole in the sky 'Everyone(i)/noone(i) was so happy he(i) jumped for joy.'
(68) a. Niemand(i) zei dat hij(i) geen tijd had. noone said that he no time had 'Noone(i) said that he(i) didn't have time.' b. *Niemand(i) zei hij(i) had geen tijd. noone said he had no time 'Noone(i) said he(i) didn't have time.' The facts from Mainland Scandinavian remain to be fully explored.
6. Conclusions and directions
Some of the initial work on embedded root phenomena - in particular the article by Hooper & Thompson (1973) - was very ambitious in its scope, aiming to provide a single unified account both for the existence and for the exact distribution of these phenomena. On the other hand, initially the data were exclusively from English, a significant limitation. Some of the major advances in our understanding of these phenomena have come from the introduction of data from other languages, although detailed comparison with data from non-Germanic languages is still required. Despite the widening of the range of languages, the more recent syntactic work has tended to leave aside issues of interpretation.
There is however an irreducibly semantic/pragmatic component to the puzzle; although the initial analyses in terms of "assertion" presented many problems of definition and explanation, no later work that has attempted to go beyond the stipulation of environments in which the various root phenomena obtain has been able to do without appeals to concepts such as factivity, assertion, presupposition, etc. A great deal of progress has been made in understanding the syntax of embedded root phenomena, in particular embedded V2: this progress has come through detailed empirical work on particular languages, and through advances in syntactic theory. However, some of the fundamental questions remain unanswered, and in fact practically unaddressed. In particular, what is the precise nature of the distinction between non-root and root clauses? As we have seen, work on V2 has overwhelmingly followed the proposal that the real distinction is between clauses with complementisers and those without. But while this proposal has been extremely productive, in one sense it simply pushes the original question one level down: why should there be such a strong (but imperfect) correlation between root clauses and empty complementiser positions?
Progress on this question is likely to come from work on the syntactic encoding of discourse semantic concepts. One version of Hooper & Thompson's (1973) proposal about the association of root clause phenomena with assertion is that root clauses have their own illocutionary force, while truly subordinated clauses do not. There is a strand of work in modern syntax that has begun to explore the syntactic encoding of such discourse semantic concepts as illocutionary force. Han 1998, citing Sadock & Zwicky 1985 and Palmer1986, points out that imperatives (morphologically marked as such) are one root phenomenon that never occurs in an embedded context. She attributes this to the existence of an imperative operator located in Comp that includes a feature for the illocutionary feature "directive", and the assumption that illocutionary force can only be expressed in unembedded clauses (pp. 150ff). Haegeman (in prep), developing proposals about the detailed substructure of the left periphery due to Rizzi 1997, proposes that there is a distinct functional head associated with illocutionary force, and that some clauses are "truncated" and therefore lack this projection, as well as others associated with e.g. topicalisation.
Much more work needs to be done in this area, both on the syntactic and the semantic side (and also on the correlated intonational distinctions). One obvious question on the semantic side is what it means for a verb complement clause (as opposed to an adverbial clause, where the question is perhaps less difficult) to be "asserted". After all, the speaker makes no commitment to the truth of the proposition expressed in the verb complement clause (in fact, as we have seen, factive verbs are typically excluded from this construction). In what sense then is it meaningful to say that these clauses express illocutionary acts of assertion? It seems that it might well be profitable to compare these cases to the "style indirect libre". There are some equally obvious questions on the syntactic side: What is the head instantiated by the complementiser that cooccurs with V2 in Mainland Scandinavian and Frisian? Is it really true that heads encoding illocutionary force cannot be embedded? If so, how can this be reconciled with the syntactic evidence for a low position for the kind of embedded V2 clauses found in German and Frisian? Answers to these and related questions have the potential for illuminating an important area in the interface between syntax and (discourse) semantics.
Return to SynCom Working Directory
Last Updated August 6, 2001 by
Rob Goedemans