Morphosyntax of verb movement and Afrikaans verbal constructions

In his Morphosyntax of verb movement: a minimalist approach to the syntax of Dutch (1997) zwart argues for an alternative analysis to the traditional analysis2 of the word order variation that exists in West Gennanic subject initial main clauses and embedded clauses. This alternative analysis is a heavily revised version of the one Zwart presented in his 1993 dissertation. The revised version focuses on a smaller section of Dutch syntax than the preceding work and revolves crucially around a proposal of feature movement and the interaction between syntax and morphology. It also deviates from the traditional analysis in the assumption that the underlying word order for Dutch is SVO and that all functional projections are head initial. Zwart (1997) furthermore claims that the analysis presented for Dutch can be carried over to the other West Germanic languages. At least one of these languages, Afrikaans, is not discussed by Zwart and my main interest in this article is to see whether Zwart's proposed analysis holds when applied to this language.

I This article is an unrevised version of an assignment submitted to the Department of Genernl Linguistics at the University of Slellenbosch in 1997 in partial fulfilment of the requirements for the MA degree in Genernl Linguistics.I wish to acknowledge the contribution of the Department and, in particular, that of Dr lohan Oosthuizen in supervising Ule work from which this article resulted.Neither the writing of this article nor my stay at the University of Stellenbosch would have been possible without the financial assistance I received from the 'Stichting YSB beurzen' in Utrecht, The Netherlands, and from the 'Stichting Neerlandia' in Bloemfontein, South Africa. 2 The traditional analysis of the verb movement in Dutch was pioneered by Koster (1975) and Den Besten (1977) among others.Two hypotheses were at the centre of this analysis.First that the underlying word order of Dutch is SOY, and second, that the Dutch finite verb moves invariably to C in tensed main clauses.view of the interaction between syntax and morphology.Finally, we will see how these proposals help to explain the word order asymmetry in subject initial main and embedded clauses and in inversion constructions.

Dutch as an SVO language
In his Morphosyntax of verb movement: a minimalist approach 10 the syntax of Dutch, Zwart argues for an alternative analysis to the traditional analysis of the word order variation in West Germanic subject initial main clauses and embedded clauses.This word order variation is exemplified in (1) for Dutch.In the subject initial main clause (la) the finite verb is in second position, whereas it is in clause final position in the embedded subject initial clause (lb).Zwart's analysis is based on the minimalist framework in general and particularly on a theory of movement and feature checking which will be presented later.He claims, firstly, that the Dutch phenomena in (I) can be profitably analysed as involving leftward movement only and, secondly, that 'a strict application of the minimalist principles leads to a simple and elegant analysis of the complicated functional domain in Dutch').In this section I will present Zwart's analysis and in the next section, consider whether it can also account for the word order variation found in Afrikaans.
Zwart's assumption of the underlying structure is based almost entirely on the restrictive approach to structure presented in Kayne (1994).Kayne proposes a theory of phrase structure ) (Zwart 1997:5) nus can only be partly true .asyet.Though the presented analysis for the verb movement asymmetry is of an elegant nature, Zwart does not apply it to constructions involving more than one verb.Its empirical merits for more elaborate verbal constructions is as yet not established.It will be shown later on in this article that Zwart's analysis in fact calls for a reanalysis of such verbal clusters in Dutch.
Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 which assumes one universal SVO underlying word order".Presenting conceptual arguments Kayne proposes that all projections look as follows': (2) All other constructions are derived from this underlying basic structure strictly by leftward movement only.No rightward movement, which in this proposal would always be lowering, is allowed.Adjunction to a head is limited to left-adjunction.

Double Agreement
In arguing for the existence of AgrS-to-C movement in the languages under discussion, Zwart (I997) starts out from the double agreement in dialects of Dutch and Frisian.He shows that all dialects that have complementiser agreement show the word order asymmetry exemplified in (1).More importantly, Frisian.spoken in the north of The Netherlands, has complementiser agreement in embedded clauses, but only if the embedded clause has not been subject to overt verb movement (Zwart 1997:198).Consider the following examples in this regard: (3) a. Heit sei datst do soks net leauwe moast dad said that-2SG you such not believe must-2SG b.Heit sei datJ"datst do moast soks net lcauwe dad said thallthat-2SG you must such not believe 'Dad said that you should not believe such things ' .
Complementiser agreement is signaIJed in these Frisian examples by the complementiser daw (with dat being the neutral complementiser).In (3a) the finite verb moast is in sentence final position which, Zwart assumes, is the verb's base position V, the object having moved to the left of this position.In (3b), however, the verb is in verb second position, to the immediate right of the subject.This is an instance of verb movement in an embedded clause.We can see from (3) that complementiser agreement and verb movement are in complementary distribution in Frisian embedded clauses.hypothesises from this observation that complementiser agreement must be AgrS-to-C movement, and that (embedded) verb movement must be V-to-AgrS movement, assuming as a standard structure the sentence structure in (4)., i.e. in PF, these bundles are spelled out as lexical items.Zwart (1997:161-7) argues that postlexical morphology and the minimalist program are highly compatible.
Zwart's proposal is the following.Lexical elements are bundles of features that are spelled out in a postsyntactic component called morphology.Throughout the syntactic derivation semantic and syntactic features are present in these bundles.Phonological features are only added after Spell-Out and play no role in the syntactic derivation.In this proposal Zwart (1997: 168-70) makes an important distinction between formal features (F -features) and lexical-categorical features (LC-features).F-features correspond to Chomsky's (1995) formal features and are those features that are involved in feature checking operations.LC-features contain semantic features and the categorial features [verbal] and [nominal].Semantic features are involved in the identification of lexical items.Although only F-features are involved in checking operations they can only be spelled out at the Spell-Out point when paired with LC-features (Zwart 1997: 182).
In the minimalist program features must be checked before the interface representations of LF and PF (Chomsky 1993(Chomsky , 1995) ) if the lexical item is to be spelled out.This checking is done by matching the features on the lexical items with features (V-features and N-features) associated with functional heads.The matching is done by movement to the functional heads.When the V-and N-features on the functional heads are strong they trigger 'overt' movement, i.e. movement before Spell-Out.Weak features only trigger movement after Spell-Out, which is in accordance with the Procrastinate principle.Another consequence of the minimalist approach is that only the formal features in the lexical feature bundle move overtly, since they are the only ones that need to be checked (Chomsky 1995:261f.)And because the overtly moved formal features cannot be interpreted in lhe morphology it is assumed that the other features follow in covert movement.
Zwart incorporates this analysis of feature movement in his analysis of verb movement asymmetry.He argues that strong V-features associated with functional heads trigger F-feature movement, or F-movement to the functional head.F-movement leaves the LC-features behind in the base position.When the verb movement is overt the presence of Fmovement alone will not be enough to spell out the verb in the higher position because the resulting structure is uninterpretable in PF.The LC-features have to move (LC-movement) overtly to join up with the F -features and to create an interpretable object.The LC-movement, in that case, is a Last Resort 7 operation.1 (See Chomsky 1993) A 'last resort' operation is applied when all other operations to avoid a derivation from crashing have been exhausted Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 The input to the morphological component mentioned above is defined as a morphosyntactic complex:.This is a head adjunction cluster containing the feature bundle.
This feature bundle is referred to as the label of the complex (cf.Chomsky 1995:243).The morphosyntactic complex is fonned in the course of the syntactic derivation by adjoining the verb features in F(v) to the tense and agreement features in T and Agr.This cluster will look more or less like (5): (5) In Government and Binding theory, as presented in (Chomsky 1981), this would correspond to the following derivation: the stem of the verb is generated in V.This stem is adjoined to T, which contains a tense affix.The result of this is a complex of V+T which in its turn adjoins to Agr, which contains an agreement affix, resulting in the structure in (5).Zwart (1997:160) refers to this as weak lericalism, which differs from his present postlexical proposal in that the verb and its affixes are already represented on the heads'.In Zwart's proposal the features instead move to assign a feature value to the features in the lexical heads.This value assignment is done in a strict sisterhood relation: the F-features (in (5» of the verb adjoin to the head T in a sisterhood relation, thus assigning the values of the verb's F -features to T. The head T in its turn passes on the F-features of the verb to its parent-node T. We say that the feature value 'percolates up' to become part of the label of the higher node (Zwart 1997:188).
This process is repeated when the parent node is adjoined to a higher head.Thus, in (6) the value of the F-features of the verb percolates up all the way to the topmost node Agr.This topmost node is the label for the entire morphosyntactic complex. (6) , zwart distinguishes the following approaches to lexicalism: weak lexicalism, which makes a distinction between derivational morphology and inflectional morphology; strong lexicalism, which does not make this distinction, and postfexicalism, which differs from the other two in assuming that the syntax manipulates lexical items generated by the morphological component of the lexicon (Zwart 1997:160).

Embedded and Main Clause Verb Movement
As we saw above Zwan hypothesises that, first, the verb moves to AgrS instead of to C and, second, the verb movement asymmetry results from an interaction of AgrS-to-C movement and V-to-AgrS movement.Feature movement is triggered by strong features on the functional heads.Zwart (1997:202) assumes that for Dutch both the N-features and the V• features of AgrS are strong.The strong N-feature accounts for the movement of the subject to the specifier position of AgrS9.The strong V-feature accounts for the movement of the F• features of the verb to AgrS.
Zwart now makes the following assumptions.Feature checking is assignment of a value, and is constrained by a sisterhood condition.This means that the syntactic relation that is expressed by the F-feature (e.g.subject-of, object-of) must be checked in a sisterhood relation.The subject must therefore adjoin to AgrSP and the F-features of the verb must adjoin to AgrS.It is in this fashion that AgrS is assigned its value: the F-features move ovenly, because the V-features of AgrS are strong, and the LC-features of the verb are left behind.This does not yet mean, as we will see, that the verb itself is actually spelled out in AgrS.
The F-movement to AgrS is not done in one step.Chomsky (1993:7) assumes AgrS and AgIO to be one Agr element.The strength on both Agr elements must therefore be equal, which means that oven movement to AgrS also implies oven movement to AgIO.Zwart (1997:203) concludes that F-movement proceeds stepwise via AgrO, and, since T also needs to be assigned a value via T as well.This is in accordance with the Head Movement Constraint of Travis (1984) which prohibits head movement across a head.Zwart assumes this constraint to be operative in his analysis.Assuming all this, F-movement to AgrS will result in the following morphosyntactic structure( 7): (7) a.
We can follow the sequence of the V-toAgrS movement by considering the structure in ( 7).
The F-features of the verb, F(v), adjoin to AgrO, assigning its feature value there, which 9 Zwart (1997:202) adopts this as a generally accepted pan of Dutch syntax and does not elaborate on it.
Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 percolates up to the directly dominating node AgrO.F(v) and AgrO, which are now in a sisterhood relation under the higher AgrO, move on and adjoin to T to assign a feature value there.The result of this second step, the topmost node T atop AgrO and T, moves to AgrS and adjoins there, creating the structure in ( 7).
Now we come to the final step in the derivation: AgrS-to-C movement.For movement to C to be justified in minimalist terms we need a feature on a lower head that needs to assign a value to a feature in the head it moves to, which, in this case, is C.
Following Den Besten (1989:92) Zwart assumes this feature in C to be Tense.According to Den Besten a clear interdependency exists between complementisers and finiteness in Dutch.
Finite embedded clauses, for instance, have also oj. or dat as a complementiser (8a), whereas a non-finite embedded clause would have the complementiser om (8b).
( The relation between the complementiser and tense is unidirectional: a particular complementiser requires a particular tense, but a particular tense does not necessarily require the presence of a particular complementiser.Zwart (1997:204) concludes that C needs to be assigned a value for tense lO , and will therefore have to attract the tense features of the verb.
These are nested in F(v).C will therefore have to attract the complex resulting from V -to-AgrS, the structure in (6).Movement of this complex to C will yield (9).We now have the morphosyntactic result of the AgrS-to-C movement which Zwart argues is characteristic of the word order asymmetry under discussion.
Up till now, however, we have only considered movement of the F-features of the verb.The complex in ( 7) is morphologically uninterpretable, since morphology can only attach a lexical item to F-features that are paired to LC-features.We saw that LC-movement is only triggered as a Last Resort operation to create a morphologically interpretable object.
Otherwise LC-movement will be covert.To explain the absence of verb movement in Dutch embedded clauses Zwart argues that in (9) the F -features combine with the LC-features in C to produce a morphologically interpretable object.The LC-features arc present in C since C is lexically filled.The complex in (9) will be spelled out as a complementiser carrying agreement inflection (overt or covert, depending on the paradigm of the language in question).The LC-features left behind in V need therefore not be moved as a Last Resort operation in ( 9).The verb is instead spelled out in V where the F-features arc still present in the form of a copy in the position of the trace resulting from F-movement (Chomsky 1993:35).This explains the word order asymmetry.
In subject initial main clauses the morphosyntactic complex does not move to C.
AgrS is the highest functional projection in these clauses and the result ofF-movement in this case is equal to ( 7).The F-features in this morphosyntactic object lack the LC-features they need to be interpreted by Morphology.The LC-features of the verb will therefore be moved as a Last Resort.They adjoin to AgrS in (7) yielding ( 10): (10) Both the F-features and the LC-features of the verb are now reunited in one object and can be spelled out by Morphology.In V we are left with two copies", one of the F-features and one of the V -features.It is unclear why Morphology does not spell out these copies as a verb.Zwart (1997:208) goes along with the general assumption that Morphology does not spell out more than one copy of LC-features and that it is the highest copy that is spelled outl2.
\I Zwart assumes that at the base position of the move<i element the trace left behind after movement is really a copy of the move<i element This is in accordance with Chomsky's copy theory of movement (Chomsky 1993 :35).

Inversion constructions
The verb movement proposal presented so far can be used to analyse inversion constructions as well.Inversion constructions differ from subject initial main clauses in that the finite verb precedes the subject instead offoUowing it (Zwart 1997 Following Den Besten (1977) Zwart concludes that the finite verb in inversion constructions occupies the same position as the complementiser in embedded clauses, i.e. in C The highest functional projection in subject initial main clauses is AgrSP.This means that there must be an extension of the tree up to CP in the case of inversion, and this extension must be triggered by a grammatical feature.This feature can either be a wh-feature, as in the case of (11 b), or a feature associated with topicalizations, as in (lla).This latter feature is referred to by Zwart as d-feature (Zwart J997:247).In both constructions in (I J) we can assume that C carries the whor d-feature respectively.These features will check corresponding features on the fronted constituent, and the feature value will then percolate up to become part of the label of the sentence as a Whole.Zwart accepts the analysis proposed in Hoekstra and Zwart (1994) that CP is in fact a combination of two non-L-related functional projections.These are WhP and TopP, and each provides a designated licensing position for wh-elements or d-words I3 , respectively.This structure will allow for the possibility of having the interrogative complementiser of and the noninterrogative complemcntiser dal combined in one construction, which is a granunatical sequence in Dutch l '.
\3 D-words are demonstrative elements often inserted between the fronted.element and the verb in topiCalisation constructions like (lla) (cf.Koster 1978): Jan (die) ken ik niet t 'John, J don " know' John that one know I not These d-words agree in ",-features with the fronted elemenl Zwart proposes that the d-feature is in fact associated.with the d-word, and not with the fronted constituent (the satellite) (Zwart 1997:248).

IDS
The difference between inversion constructions and subject initial main clauses now follows from the analysis of verb movement advanced by Zwart.Recall that when C is present, as we assumed for inversion constructions, the morphsyntactic complex resulting from V-to-AgrS moves on to C, yielding the morphosyntactic object illustrated in (9).lb.is movement to C is again, as in embedded constructions, triggered by the need for C to be assigned a tense feature by T. Zwart now argues that contrary to embedded constructions, C in inversion constructions does not contain the LC-features of the complementiser.which makes the morphosyntactic object in (9) uninterpretable for morphology.lb.is triggers a Last Resort movement of the LC-fcatures of the verb, which yields the following structure: This structure will be spelled out by morphology as a verb, which explains the inversion of subject and the finite verb.This completes our survey of the main points raised in Zwart (1997).We will now tum our attention to Afrikaans.

A look at Afrikaans
Now that we have an idea of what Zwart's proposal entails we will consider whether it correctly predicts the grarnmaticality of all Afrikaans verbal constructions.We might expect this to be the case, since in ch.VI.5 Zwart puts forward the hypothesis that AgrS-to-C movement explains verb movement asymmetry in all Germanic languages that display it, and, conversely, that the absence of such an asymmetry ought to follow from the lack of AgrS-to-C movement.This claim predicts that since Afrikaans is a Germanic language, IS the verb movement phenomena in Afrikaans are also a result ofV-to-AgrS and AgrS-to-C movement.
Let us follow Zwart's argumentation for the applicability of the verb movement analysis to the other Germanic languages and see whether the same indeed holds for Afrikaans.
I' I ignore here the Wlfesalved and ongoing debate about the 'status aparte' of Mrikaans in the sense that its creoUde histor)" eroded its Germanic character to such an extend a!l to disqualliY it as a purely Germanic language (for this matter see: Den Besten 1986, 1989, Raidt 1983).In this paper I will treat Mrikaans as a continental West Germanic language like its 'parent language' Dutch.

I Afrikaans as a Germanic language
Recall that Zwart argues that the presence of AgrS-to-C movement is signalled morphologically as complementiser agreement in several languages and dialects, including Frisian, as exemplified in (3).The fact that Frisian only displays complementiser agreement in embedded clauses when there is no overt verb movement is indicative of a complementary distribution of complementiser agreement and verb movement (Zwart 1997: 198).From this Zwart makes the generalisation that all languages that lack cmbedded verb movement must have AgrS-to-C movement, even though they might lack a complementiser agreement paradigm to show for it.It is assumed that the absence of such a paradigm is a superficial phenomenon and does not in fact mean that this would affect the syntax of such dialects or languages.The asswnption is therefore made that abstract complementiser agreement is present in these languages.
Two preconditions for the presence of overt complementiser agreement in a language were formulated in Hoekstra (1992).The first precondition is that the nominal plural forms and the verbal plural forms must be identical, and the second precondition is that there must be a morphological opposition between singular and plural forms in the verbal paradigm.
These preconditions are not met for Afrikaans: although Afrikaans has plural inflection for nouns, the verbal paradigm lacks such an inflection, neither is there a morphological opposition between singular and plural in the verbal paradigm.This paradigm, in fact, contains only one form for all persons and numbers.This rightly predicts the absence of overt complementiser agreement in Afrikaans.This does not automatically mean that Afrikaans has covert complementiser agreement instead.One could therefore not conclude that AgrS-to-C takes place in Afrikaans solely because 'the AgrS-to-C hypothesis provides a satisfactory account of the verb movement asymmetry in this language', although Zwart does so for the Mainland Scandinavian-Frisian languages (Zwart 1997:230).The generalisation that AgrS-to-C movement takes place in all Mainland Scandinavian-Frisian languages is based entirely on observations from the other Germanic languages but has no direct empirical justification in data from Afrikaans 16 We will see as well that unlike Dutch, Afrikaans is not a clear member of the Mainland Scandinavian-Frisian group of languages, since it does not share all the characteristics of that group.For the moment, however, we will go along with Zwart's argumentation.
Afrikaans in general lacks verb movement in embedded clauses, and, in the overriding number of cases behaves the same as Dutch, the West Germanic language it descends from.In her dissertation Robbers points out that for these contexts the situation is rather different for Afrikaans (Robbers 1997:28-32).Not all the contexts mentioned in ( 15 The sentence in ( 17) is reminiscent of the bridge verb constructions mentioned above (cf. « 16)a, (I6)b).The difference here is that the complementiser introducing the embedded clause is not overt, or even totally absent.These sentences can easily be analysed by assuming that the subordinate clause does not project up to the CP level.In that respect, it can be regarded as a normal subject initial main clause with an AgrSP as its highest functional projection.There is no C to attract T and thus trigger AgrS-to-C movement.The LC-features will be moved to AgrS as a last resort and the finite verb is spelled out in AgrS.
Embedded verb movement also occurs in embedded clauses introduced by a complementiser: Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 (18) Sy filosofie is dat ons aanvaar dinge his philosophy is that we accept things 'His philosophy is that we should accept things' Zwart's analysis would predict that in cases such as ( 18) no overt AgrS-to-C takes place, which would again trigger last resort movement of the LC-features of the verb.Zwart (l997:233ff.)argues that AgrS-to-C movement is the process that lifts the barrier status of the complement, i.e. the clause is made transparent.Absence of AgrS-to-C therefore would make the embedded clause non-transparent.Zwart assumes that in Yiddish and Icelandic AgrS does not move to C in overt syntax, but in covert syntax.For Mainland Scandinavian-Frisian Zwart assumes that in embedded verb second clauses AgrS moves neither in overt nor in covert syntax.The obvious question is why in these cases C need not attract T through AgrS-to-C.
Zwart claims that the lack of this need to anract T should be ascribed to the contexts that allow or do not allow embedded verb movement in Mainland Scandinavian-Frisian.
These are the contexts exemplified in ( 15) and ( 16) above.Zwart (1997:236-7) argues that in these contexts there is no need for C to anract T because the embedded clause is a root clause, i.e. it has the characteristics of an independent main clause, though introduced by a complementiser and functioning as an embedded clause.This does not explain, however, why some of the same contexts that do not allow verb second in Mainland Scandinavian-Frisian do allow verb second in Afrikaans (cf.( 15) and ( 16)).Considering this issue and recalling the lack of complementiser agreement in Afrikaans it would seem that it is impossible to establish independently whether C does ever attract T in Afrikaans.
Embedded verb second in Afrikaans is, however, more diversified than would appear from ( 17) and ( 18).Besides the root phenomena or bridge verb contexts there are a number of other embedded verb second situations that apparently do not occur in the Mainland Scandinavian-Frisian type languages.Consider the following Afrikaans embedded verb second clauses taken from various sources (Fonelis 1993, Feinauer 1989, Oosthuizen 1996) Zwart suggests the following description of embedded verb movement (Zwart 1997:23\).As in subject initial main clause verb movement constructions and subject initial embedded clauses both the N-features and the V-features of AgrS are strong.This triggers movement of the subject to the specifier position of AgrS and movement of the F -features of the verb to the head position of AgrS ' 9. Contrary to subject initial embedded clauses with the finite verb in final position, however, T is not attracted by C and the morphosyntactic complex will remain in AgrS.So will the F-features of the verb.No overt AgrS-to-C movement takes place.The LC-features of the verb are therefore forced to move as Last Resort to provide the stranded F-features in AgrS with a morphologically interpretable object as they would do in subject initial main clauses.As a result, the verb is spelled out in AgrS, the verb second position, and we havc a verb second embedded clause as a result.
The embedded questions in ((I9)a -(I9)d) are all instances of embedded inversion constructions.In each of the embedded clauses the finite verb precedes the subject.Zwart's analysis of inversion constructions would predict that due to the inversion the finitc verb is expected to be in C, but the fact that the clause is embedded would have to result in the complcmentiser being spelled out in C. The question naturally presents itself whether in these situations T is attracted by C. Zwart's analysis predicts that it is not possible to embed an Ik weet niet of dat er werk zal zijn I know not whether that it work shaU be '1 don't know whether there wiiJ be work' 'This is possible for Dutch, as shown by ( 21), but the same is not grammatical in Afrikaans, as is evident in (22).( 22) -Ek weet nie of dat daar werk sal wees nie I know not whether that there work shall be NEG '1 don't know whether there will be work' 'This sheds some doubt on the existence of the 'double' CP for Afrikaans. Oosthuizen (1996:92), however, considers the combination of an interrogative and a non-interrogative complementiser a possibility in non-standard forms of Afrikaans.The complementiser dat can only be spelled out in the head position of TopP when no inversion takes place in the embedded clause (cf.sentence (l6)a)'0: 10 The grammaticallty judgements that I gathered indicate that a sequence of wh-words like wat, dot, waar and the complementiser dot is granunatical in Afrikaans (cl.(23a», but that the sequence of dat is not grammatical.This would suggest that we arc in fact dealing with the single lexical item afdat, which, though a Dutch lexical item, is not part of the lexicon of Afrikaans.This would be in contradiction with the argument in Hoekstra and Zwart (1994) that the Dutch of dol comprises two lexical items and not one.They try to show this by applying conjunction reduction (Hoekstra and Zwart 1994:193): Hoekstra and Zwan' argue that lexical items that form one head and not two cannot be split in a conjunction construction.In (a) the second prefix vercannot be deleted.This proves, it is argued, that the verbs vermaken and verknippen arc one lexical head and cannot be split.The argument is similar for the complementiser of dal: Zwart and Hoekstra argue that because this complcmentiser is in fact a combination of two lexical heads it is possible to delete the first part in the second clause of the conjuction construction, as is shown in (e).This will not produce an ungrammatical construction.What Hoekstra and Zwan fail to point out, though, is the fact that in (c) we might just as well be dealing with two different lexical items, one being afdat and the other dat, each introducing one of the conjuncted clauses.The conjunction constructions above, therefore, can hardly be seen as convincing evidence for the existence of two lexical heads in afdar.The ungramrnaticality of Afrikaans afdat in ( 22) would moreover suggest that this is not an item in the Afrikaans lexicon, although the separate afand dot are.This complementary distribution of the complementiser and the inverted finite verb in (19)a and ( 23) suggests that both are spelled out in the head position of TOpp21.If we were to assume the existence of a double CP for Afrikaans, however, we would still not be able to explain why a sentence like ( 24) is also possible in Afrikaans (Oosthuizen 1994: 167):

Stellenbosch
(24) My gevoel is dat het dit maar so gebeur!my feeling is that has this but so happen My feeling is that if only things had happened that way)' So far the Afrikaans embedded inversion constructions we have seen here introduced by a wh-word like oj, but the embedded inversion construction in ( 24) is introduced by a noninterrogative complementiser dat.Even when we assume the existence of the double complementiser position we cannot accommodate both the complementiser and the inverted finite verb.We would have to assume both to be positioned in the head position of TopP.This we cannot do.
Neither would we be able to explain why the embedded inversion in (19c) is allowed in Afrikaans but not in Dutch while we still assume the same underlying structure.This 21 Den Besten (1989: 160) even mentions the existence of the sequence of a wh-word, a complementiser and an inverted finite verb: ... hoekom dat het hy dit gedoen why that bas be it done .... why he has done it' This construction, however, is rare in Afrikaans and is considered 'very low', according to Den Besten.I will leave it out of consideration in this discussion.
The last instance of embedded verb movement in Afrikaans, the relative clause as exemplified in (lge) and (l9f) does not present any real problem.The analysis here is similar to that presented for (18), the clause introduced by a comp/ementiser with the finite verb in AgrS.We can similarly assume here that the relative complementiser wat in both sentences does not attract the Tense features in AgrS.AGRS-to-C will not take place, the LC-features of the verb move as a Last Resort and the [mite verb is spelled out in AgrS.

Open finals
To round off the survey of embedded verb second in Afrikaans we will have a cursory look at a phenomenon related to embedded verb movement constructions.Recall the as yet undiscussed Afrikaans sentences in (l9)h and (I9)i.This Afrikaans construction is unknown in Dutch and is presented as an open final construction in Ponelis (1993:331).In this construction the verbal cluster at the end of an Afrikaans sentence can be intenupted by the object or other non-verbal material.This is not possible in Dutch.Both of Zwart's earlier proposals for the analysis of verbal clusters suffer from one significant drawback when incorporated in the verb movement analysis in Zwart (1997) All the movement operations suggested in both Zwart (1993) and Zwart (1995) are incompatible with the principles of the general base hypothesis adopted in Zwart (1997) In this hypothesis movement must be triggered by a feature checking requirement and verb movement must be head-to-head movement.Consider the morphological structure of the participle: consistency would demand that similar to the finite verb, a morphosyntactic structure functions as the input to the morphological component for the past participle as well.This means that there must be F-feature movement and LC-movement for the participle as well.One could think of a [i perfect] feature as a trigger for the licensing of the past participle.
The point that needs to be made in connection with the subject of this article is that Zwart's analysis might account for word order asymmetry but also triggers the need for a reevaluation of tbe analyses proposed so far for the more complex verbal clusters.One would also have to consider the consequences for Zwart's verb movement proposal when introducing a new structure, like the PredP introduced in Zwart (1993).What would be the consequences for the morphosyntactic complex resulting from F-movcment? Would it include " Zwart argues that if participles are licensed in specifier positions, and specifier positions arc always to the letl, there is no way in which a cluster consisting of an auxiliary find participle (e.g.Dutch heeft gewerkl, 'hIlS worked') could be derived starting from a head fmal structure.Zwart considers this to be a strong support for the assumption of a head initial structure as the basic structure for multi-vcrb constructions in all Gennanic languages (Zwart  1995:225).Considering Afrikaans however, we can argue for an el opposilum: the same word order is ungrammatical in Afrikaans verb clusters (i.c.'hel gewerk, 'has worked' does not occur in embedded clauses) Exccpt for a mnin clause sentence like ek het gewerk, 'I havc worked' (which is a sentence that is explained by the verb movement analysis in Zwart 1997), the auxiliary will always follow the partiCiple in Afrikaans.Combining this "ith the fact that this order CJlMot be derived starting from a head fmnl structure can equally well be argued to be 'strong support' for this head fmal structure.

Conclusion
This article has attempted to apply tne analysis of verb movement and feature movement proposed in Zwart (1997) on data from Afrikaans.Zwart's proposal has,an initial appeal for the analysis of Afrikaans since it accounts for the general asymmetric word order constructions that resemble those in Dutch, However, a number of instances of Afrikaans embedded verb second defy this analysis.
Afrikaans cannot be classified in the same way as other Germanic languages.
Especially on the contexts in which embedded movement is allowed or not allowed in Mainland Scandinavian-Frisian languages, Afrikaans differs sharply.This casts doubt on the assumption that it is through the lack of a need for C to attract T that embedded clauses remain non-transparent and thus allow for embedded verb movement.
Zwart's analysis does not predict the fact that Afrikaans allows for embedded inversion constructions.Though this problem might be solved by assuming a double CP, this would not explain why the same embedded inversion constructions do not occur in Dutch.
Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 leest het boek Peter reads the book 'Peter reads the book' b. ik zie dat Peter het boek leest I see that Peter the book reads '] see that Peter reads Ihe book' the pivot of verb movement in West Germanic languages, and not C, as was the case in the traditional analysis•.Zwart then argues that not only Frisian but all complementiser agreement dialects have verb movement asymmetry (d (I».1bis syntactic generalisation can be made without exception for aU overt complementiser agreement morphology.Although Standard Dutch itself does not have overt complementiser agreement morphology Zwart concludes that it has overt AgrS-to-C movement as well, asswning that it only lacks a panuligm of inflected complementisers to show for it.This leads to the assumption that Standard Dutch does have abstract complementiser agreement.2.3 Feature Movement and the interface between syntax and morphologyIn addition to the assumption of AgrS as the pivot of verb movement in Dutch Zwart advances a proposal of feature movement(Zwart 1997:ch V).He bases this proposal on the definition of Feature Movement in Chornksy (1995:262) and on the concept of Distributed Morphology introduced by Halle and Marantz (1993).Chomsky's concept of feature movement was advanced as the most economical way to check formal features, that is, those features involved in feature checking operations (tense, agreement and case, for example).Halle and Marantz' Distributed Morphology, which is referred to as pos/lexica/ism by Zwart, defines the relation between morphology and syntax.In their view, stems and morphemes are only bundles of formal and semantic features during the syntactic derivation.In postsyntactic • In the lraditional analysis, Dutch was assumed to have an underlying SOY word order.The finite verb was assumed to move to the COMP, or C, position through V-to-C movement (Den Besten 1977, Koster 1975).Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 " E.g. a sentence like: 'Jan vroeg of dat ik haar kende' John asked whether that r her knew Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58

19
Zwart leaves it open whether or not movement of the subject and object involves separate F-and Lemovemellt as well.Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 in version construction.Clearly Zwart's analysis makes the wrong prediction here.Let's see ifwe can find a way out.Consider the structure underlying sentence (19)c as illustrated in (20): of the embedded clause is represented.In this clause the subject hy moves to the specifier position of AgrSP, triggered by the strong N-features on AgrS.The subject is thus in a position where it can be licensed, i.e. in a sisterhood relation with the projection of the functional head AgrS.C contains the Fand LC-features for the complementiser of The F-features of the finite verb sal move to AgrS as V-to-AgrS movement.Note now that if we assume that inversion takes place in the embedded clause, this would mean that movement of the morphosyntactic complex in AgrS would have to be triggered by the lack of LC-features in C. In our example (I9c), however, the presence of the complementiser of means that C is furnished with LC-features, thus blocking inversion.Neitber could we have the complementiser and the finite verb being spelled out in the same position.We would need an alternative position between C and AgrS for the inverted finite verb to be spelled out.One way out might be to assume that CP is in fact a composite of the two functional projections WhP and TopP, as suggested earlier.The interrogative complementiser of in that case would have to find itself in the head position of WbP and the inverted verb would have to be spelled out in TopP.According to Zwart strong support for this division of CP into WhP and TopP is provided by the fact that in Dutch the interrogative complementiser of and the non interrogative complementiser dal can be combined in one construction: Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58(21) ver-maakt en ver-knipt -De jurk is ver-[maalct en -kniptJ b.Hoewel hij jong is en hoewel hij weinig eet -hoe-[wel hij jong is en -wei hij weinig eetl c. 1k wed waarom of-datje komt en ofdatje zo vroeg weer weggaat Ik weet waarom of-ldatje komt en -datje zo vroeg weer weggaatl Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 lIS (23) a. Ek wil weet wat dat hy doen.b. *Wat dat doen hy? c. *Ek wil weet wat dat doen hy.
is a clear problem to which I have no immediate answer, although one might consider the possibility of maintaining a single CP besides the presence of a double CP consisting of a WhP and a TopP.One would in such a case have three positions availabe: WhP, TopP and CP.In the case of (24), the inverted finite verb would move to CP and the complementiser would be spelled out in TopP.The difference between Dutch and Afrikaans could then possibly be explained by assuming that the extra CP is a feature solely of Afrikaans, and not of Dutch.This assumption of an extra cp 12 , however, demands extra research beyond the scope of this article.
speakers of Afrikaans.Their occurrence varies widely amongst the different varieties of the language, but the construction has a long standing history in the development of Afrikaans.(See Ponelis (1993) for more details.)Notice that when the open final occurs in an embedded construction,.as in «25)e -(25)g), the finite verb may appear to be in second position.There is, however, a crucial difference between open finals and embedded verb second clauses.Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58This analysis of verbal clusters is abandoned in Zwart (1995) in favour of another analysis.In that work Zwart formulates two processes that arc assumed to cause word order variation in Continental West Germanic (Zwart 1995:216): (29) Movement in verbal clusters: a. Adjunction of an infinitival verb to a modal verb (Xo-movement).b.Raising of a participle to the specifier position of an auxiliary verb (XPmovement)25.Zwart (1995) argues that infinitives undergo head movement and arc licensed by adjunction to the immediately higher verb.Participles undergo XP-movement and are licensed by movement to the specifier position of an auxiliary verb.The latter process can be extended by moving the participle to the specifier position of a modal verb which an infinitival auxiliary has been adjoined to (see Zwart 1995 for details).
Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 movement to the new XP, i.e. would the head Movement Constraint (Travis 1984) be observed?Acceptance of Zwart's proposal clearly triggers the need for the re-evaluation of much of the data involving verbal structures with more than just a simple finite verb.Zwart's proposal would thus appear to have very little empirical value outside the simple present verbal constructions, and to run the danger of triggering the need to rethink more than it initially explained.Real success for Zwart's analysis of the word order asyrnmerty is only achieved through its incorporation into a broader analysis of the more complex and intricate verbal workings of the languages it claims to clarifY.
Hooper andsentences in (I), for instaKroch 1992)ate without any change in word CVVe will see presently whether these are the same contexts as the embedded verb movement contexts in Mrikaans.)Themostimportant context in Mainland Scandinavian seems to be in the complement of bridge verbs,Hooper and Thompson 1973, laaidou andKroch 1992).Put more simply, the embedded clause or AgrSP qualifies as a full main clause, although it is introduced by a complementiser.Zwart argues as follows.We can assume that AgrS-to-C movement does not take place in these assertive embedded clauses.The reason for this is that the embedded clause need not be marked as dependent on the matrix clause.AgrS-to-C movement would mark AgrS as dependent on C (by assigning a tense value to C).
Vikner (1995)eter reads the book' Following Zwart, we would conclude that Afrikaans has abstract complementiser agreement and therefore lacks AgrS-to-C movement in embedded clauses.There are, however, a number of cases in which Afrikaans has embedded verb second, which makes classification of Afrikaans in the Germanic language group problematic.But before we have a look at some concrete examples of embedded verb second we will first consider Zwart's classification of embedded verb second languages.FollowingVikner (1995)Zwart distinguishes two types of embedded verb movement in Gennanic languages.One is the Yiddish-Icelandic t)pe, which consistently moves verbs in all embedded clauses, and the other is the Mainland Scandinavian-Frisian 17 type, which only moves the embedded finite verb in specific circumstances.With embedded verb movement occurring only in certain situations Afrikaans would thus, at first glance, best fit in with the Mainland Scandinavian-Frisian type.Yiddish-Icelandic is described as a type that never has AgrS-to-C movement in overt syntax.The Mainland Scandinavian-Frisian type, in contra.st,doeshaveAgrS-to-Cmovement in overt syntax.exceptwhereverbmovementtakesplace.For Mainland Scandinavian and Frisian Zwart (l997:234ff.)distinguishesthefollowingcontexts in which embedded verb movement takes place.Bridge verb constructions are basically the contexts in which English allows embedded root phenomena(Zwart 1997:235).Embedded clauses with root phenomena contain the assertion \ 1 The Mainland Scandinavian languages are Swedish, Danish and Norwegian.Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 of the sentence as a whole and, contrary to other, 'nonnal', sentences, where the assertion is contained in the matrix clause, this assertion is found in the embedded clause" (cf.'Pete said that he did not know that book' b.Hij kende dat boek met, zei Piet He knew tbat book not said Pete 'lJe did not know that book, Pete said'(The example in a. is provided by Zwart, and is claimed to be colloquial Dutch.To my ear, however, the grammaticality of this construction seems highly doubtful).(See Hooper and Thompson 1973 for more on root phenomena).Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 g.Oat Jan clat boek kent is verrassend that John that book knows is surprising 'That John knows that book is surprising' Zwart (1997)ubjects, finally, do not aUow embedded verb movement in Afrikaans, as is evident from « 16)d).We see now that Afrikaans agrees with Mainland Scandinavian-Frisian on only one of the four embedded verb movement situations illustrated.To summarise, Afrikaans does not meet Hoekstra's preconditions for the presence of overt complementiser agreement, and should therefore have, according toZwart (1997), covert AgrS-to-C movement.The presence of AgrS-to-C in Afrikaans would qualify it as a member of the Mainland Scandinavian-Frisian group of Germanic languages.However, Afrikaans differs markedly on the contexts in which embedded verb movement is allowed.This, however, need not necessarily hamper the applicability of the analysis on Afrikaans verbal constructions.So let us now look at the actual Afrikaans embedded verb second clauses and see whether they are correctly predicted by Zwart's analysis.
As we saw in (15) embedded verb movement is not allowed in Mainland Scandinavian in the complement of negative and negated verbs, in irrealis complements, adjunct clauses and sentential subjects.In (16) a rather different picture emerges.Verb movement in the complement of negative verbs is allowed in Afrikaans «16)a).So is verb movement in irrealis Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 contexts «(I6)b).Verb movement in adjunct clauses is also widely used, especially when introduced by dal-type complementisers like dol 'that'.noudal'nowthat'.sodal'inorderto', etc., but also by other complementisers like as in « 16)c).3.2EmbeddedverbsecondEmbedded verb second constructions have long been attested in Afrikaans and are frequently mentioned in the literature.It is especially common to find embedded verb movement in clauses that are not introduced by a complementiser.Consider for instance (17):'/ know he has read the book' . Dit is vir die dag wat jy kan iets wegsit this is for the day that you can something put-aside This for Ihe day Ihal YOll 'Il be able 10 put something aside' f.Dis by onse plaas wat my pa het vir hulle gese, hulle moet loop this-is at our farm that my dad has to them said they must walk '11 was at our farm that our dad told them 10 go away' g ... dat sy graag sal die bock wil lees ... that she eagerly will the book want read ' ... that she will eagerly want to read the book' h ... dat sy nie sal die boek wil lees nie ... thal she not will the book want rcad NEG ' ... that she won 'I wanllo read Ihe book' (19)a and (19)b arc instances of embedded wh-clauses with verb movement.Besides whquestions.yes-no questions can also appear as embedded verb movement clauses, as in (19)c and (J 9)d.(J 9)e and (19)f arc examples of relative clauses with verb movement and (19)g and (l9)h, finally, are instances of open finals, which, as we will see further on, differ from [ have wondered whether will he come '} have wondered whether he will come' d.Ek weet nie of kan die bos miskien weer lewendig word rue [ know not if can the forest maybe again alive become NEG '} do not know whether the forest could become revived again' Stellenbosch Papers in Linguistics, Vol.31, 1998, 95-124 doi: 10.5774/31-0-58 eembedded verb second clauses but arc not predicted by Zwart's analysis.Let's discuss the sentences in ( 19) one by one.