Oblique Case and Subjecthood or: Why Icelandic Is Different

  • Eric Haeberli
Part of the Studies in Natural Language and Linguistic Theory book series (SNLT, volume 54)


In chapter 3, I outlined a theoretical approach to the traditional observation that there is a correlation between the presence of a morphological case system and the occurrence of relatively free argument order. In the second part of chapter 4, I proposed an analysis of the cross-linguistic variation found among the Germanic languages with respect to the occurrence of adjuncts in pre-subject position. Finally, a part of chapter 5 dealt with variation concerning the co-occurrence of expletives with definite subjects in the Germanic languages. What is common to all these analyses is that they identify one language in the Germanic language family whose behavior is unexpected, namely Icelandic.


Categorial Feature Case Feature Nominative Object Clause Structure Dative Argument 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    Cf. also e.g. Shlonsky (1989) and Ritter (1995) for such a distinction in the clause structure of Hebrew. Their proposal is related to the fact that Hebrew exhibits contrasts among tenses with respect to whether person and number (past, future) or only number (present) is morphologically realized as an agreement affix on the verb.Google Scholar
  2. 2.
    Cf. also Ritter (1995) and Shlonsky (1989) for Hebrew. Both of these authors also conclude that person is above TP and number below TP.Google Scholar
  3. 3.
    Cf. also Davis (1998), Alexiadou (1999) for associating person to tense in entirely different contexts.Google Scholar
  4. 4.
    There may be an alternative analysis for the statusof AgrN however. In chapter 3 fn. 21,1 pointed out that, like the VP, the inflectional domain may also contain more than one projection bearing the same categorial features, i.e. that there may be a recursion of inflectional projections. Hence, it could be argued that AgrN is related to a lower inflectional projection (e.g. for aspect) whereas Agrp is related to a higher inflectional head (tense).Google Scholar
  5. 5.
    An additional issue that might arise in connection with the assumption that agreement features in Icelandic are generated below T and on T is the morphological representation of these features on verbs. A traditional assumption since Baker’s (1988) work has been that the order of morphemes is a reflex of the syntactic structure (Mirror Principle). Thus, it has been assumed that the morpheme that is attached directly to the verb is associated to the projection right above the VP in the clause structure and that subsequent morphemes are associated to higher projections. In other words, the closer a morpheme is to the verb the lower its associated projection is in the clause structure. Applying this type of reasoning to the paradigm in (11) would therefore suggest that, since the order of morphemes is T-AgrN-Agrp, both Agr projections are above TP, contrary to what is suggested in the text. I therefore propose that the morphological sequence T-AgrN-Agrp is simply a reordering effect which occurs within the morphological component (i.e. post-syntactically). The motivation for this effect could be argued to be the fact that the two agreement features are directly related in the sense that they are two aspects of an agreement relation with the same element. Thus, the morphological component groups the two elements together although in general the morphemes occur in the order in which they appear on the complex head after V-movement.Google Scholar
  6. 6.
    Note that in the example shown in (16) T would be the only possible host for the DAT feature if we assume that heads which generally do not assign case such as participial T or auxiliary V do not have a case feature slot in their feature matrices. However, this does not mean that T is always the only option for inserting a hostless DAT feature. Another option is illustrated in (i) (from Sigurðsson 1992:8). (i) Við töldum beim hafa verið hjálpað Icelandic We believed them (DAT) have been helped ‘We believed them to have been helped.’ (i) illustrates an oblique subject construction in combination with an ECM verb. The ECM verb in (i) generally assigns ACC to the subject of the embedded clause and it therefore generally bears an ACC feature. By analogy to the proposal made for matrix T, we may assume however that this feature is only optional and that therefore in (i) the host-less DAT feature can be inserted under matrix V instead of ACC. Hence, the subject of the embedded clause bears DAT in (i) rather than ACC.Google Scholar
  7. 7.
    It should be pointed out that I have been simplifying somewhat one aspect of the derivation of (15). Participles also can be inflected and forms like hjálpad which are used in passive constructions lacking a Nominative argument correspond to the Nominative singular neuter form (cf. e.g. Thráinsson 1994:177). Given the assumption that agreement properties have to be checked at least once in a local configuration, this specification of the participle would also have to be checked. The non-overt expletive proposed in the text would therefore possibly have to be inserted in the participial clause already. However, lower insertion of the expletive would not lead to a change in the crucial aspects of the derivation discussed in the text. In particular, the oblique argument would still end up in the highest inflectional position thereby obtaining the status of subject. By neglecting participle agreement, I therefore have simply avoided additional complications which would not be directly relevant for the main point made in the text.Google Scholar
  8. 8.
    This assumption is not crucial however. Alternatively, we could say that the empty P-head of the DAT argument can be incorporated into the next higher head and that PP-transparency then allows T to select and attract the DAT argument for checking. An expletive would then have to be inserted in the case checking projection above TP and then it would move to the structural subject position in AgrsP. The result with respect to subjecthood would thus be the same as in the approach discussed in the text.Google Scholar
  9. 9.
    An important question that arises in this context is whether the Icelandic system is the only one being able to derive oblique subjects in the framework proposed here, i.e. whether a language with oblique subjects has to make a person/number agreement distinction and have independent case features. The second property may indeed be important for obtaining oblique subjects (cf. also fn. 20 below). However, the first property does not seem to be a prerequisite for oblique subjects because there are alternative options allowing non-Nominative elements to occur in the highest inflectional projection. Consider again for example Korean. Korean has constructions in which a nominal constituent marked with Dative seems to have the status of a subject and a nominal constituent marked with Nominative is an object (cf. e.g. Lee 1993:75, Lee 2001:71). Since Korean does not make a morphological person/number agreement distinction, it would not be very attractive to relate the presence of oblique subjects to a syntactic person/number distinction. However, I proposed in chapter 4 (fn. 6) that AgrP above TP can be motivated in Korean on the basis of the occurrence of honorific agreement morphology. What is interesting for our purposes now is that Dative subjects are able to trigger honorific agreement (cf. O’Grady 1991:102, Lee 2001:71). In terms of our analysis, this suggests that, in contrast to German or Icelandic, Dative nominal constituents in Korean are not structurally incompatible with agreement and they therefore enter a licensing configuration with Agr above TP if they are subjects (either via an empty expletive or through overt movement). Dative arguments can therefore be associated with the highest inflectional specifier position in Korean even though no person/number agreement distinction is made. Future research based on detailed cross-linguistic evidence on oblique subjecthood will have to determine more precisely what options there are for deriving oblique subjects within the framework outlined here. For example, it would be conceivable that oblique subjects can also occur in contexts in which no agreement projection occurs at all in the inflectional domain, but 1 have to leave it open here whether this option is realized in some languages. However, the case of Korean suggests that, given certain distinct morphological properties, oblique subjecthood can indeed be derived in other ways than described for Icelandic in the text.Google Scholar
  10. 10.
    The 1st and 3rd person agreement endings are identical in the past tense in Icelandic. Hence, (21a) could be interpreted either as involving agreement triggered by the 1st person pronoun or as default agreement. Given the slightly less degraded status of (21a) as compared to (21b/d/e) and given the observations made in the context of (23) below, the latter option seems to influence the speakers’ intuitions concerning (21a). The judgement given in (21a) is Sigurðsson’s, but as (23) will show there is considerable variation among speakers here.Google Scholar
  11. 11.
    Cf. the discussion after example (24) for more details on AgrN checking.Google Scholar
  12. 12.
    One additional aspect of this issue is the following. There are raising verbs in Icelandic which allow the subject of their complement to occur in the Nominative case if a Dative argument is the subject of the matrix clause. This is shown in (i) (from Sigurðsson 1996:30). (i) Mér virtust / virtist Þær vinna vel Icelandic Me (DAT) seemed-3PL /seemed-3SG they work well ‘They seem to me to work well.’ What is interesting with these constructions is that they generally tolerate 1st and 2nd person Nominative arguments with default agreement on the verb. (ii) Þeim hefur alltaf fundist við vinna vel Icelandic Them (DAT) has-3SG always found we (NOM) work well ‘They have always thought that we work well.’ [continued on next page] The contrast between (ii) and (23) is surprising. However, we may relate this contrast to the suggestion made in the text. If the problem with (23) is indeed one which is related to the licensing conditions of the pronoun, then we could account for the contrast between (ii) and (23) in terms of the different syntactic status of the Nominative element in the two constructions. In (ii), the Nominative element is the subject of the infinitival clause. What we could argue then is that the licensing requirements of a lst/2nd person pronoun only concern the options which are available within the clause that contains the pronoun. An infinitival clause does not have agreement and we therefore could argue that agreement is irrelevant for the lst/2nd person pronoun in (ii). Hence, it can occur in a context with default agreement. In (23) however, agreement is not related to a different clause but to the clause to which the Nominative element belongs as one of its arguments. Hence, the pronoun should be licensed by the agreement features available in its clause.Google Scholar
  13. 13.
    But cf. section 5 below for a difference with respect to case checking which, however, does not have an influence on the syntax of subjects.Google Scholar
  14. 14.
    As already pointed out in section 2.1.6, there is one context in which Agrp has to project in order to create a specifier position, namely in cases where Agrp does not have a default value but instead bears 1st or 2nd person features. But this scenario is restricted to pronouns and pronouns are generally adjacent to C even in those languages which allow non-adjacency with full DP subjects (cf. e.g. den Besten 1983, Platzack 1986 and Vikner 1995:44f, 103f.). This adjacency requirement suggests that pronouns obligatorily move to agreement positions to be licensed. But if pronominal subjects move for agreement licensing, then the occurrence of a [Spec, AgrpP] does not change anything with respect to C-subject adjacency because the highest specifier position (i.e. the position in which Agrp is licensed) will always be occupied by the pronoun. Thus, the conclusion that AgrpP sometimes projects a specifier position does not lead to the wrong prediction that C-subject non-adjacency should be possible in this context.Google Scholar
  15. 15.
    An alternative would be to say that the subject occupies an independent position where it is assigned contrastive stress, e.g. a specifier position of a focus projection which occurs between TP and the Nominative case checking position. Cf. also Laenzlinger (1998:289) for the claim that focalization can lead to non-overt movement for case checking.Google Scholar
  16. 16.
    A brief remark is necessary here concerning the licensing of the empty expletive. In chapter 5 (example 3), I proposed that empty expletives are licensed in a local configuration with Agr. But in the derivation described in the text the empty expletive occurs in a case checking position rather than in an agreement licensing position. Two options are available here. First, it could be argued that empty expletives in Icelandic are licensed by Agr through cliticization to the C-head where Agrp licensing takes place, i.e. that the licensing configuration in these cases is a head-head structure. An alternative would be to assume that, as suggested earlier (cf. section 2.1.6), Agrp is a subfeature of case on T in Icelandic and that, due to this close relation between agreement and case, an empty expletive can be licensed in a case checking position.Google Scholar
  17. 17.
    Note that the occurrence of AgrN P and hence of AgrN checking in the clause structure of Icelandic does not have any influence on the conclusions drawn in this paragraph. As proposed in section 2.1.6 above, AgrN is not an attractor. It therefore does not have the capacity of moving an expletive away from its associate.Google Scholar
  18. 18.
    However, as observed in chapter 5 (example 32), definite subjects and expletives can occasionally co-occur in Icelandic if the predicate is an ergative verb. The analysis that I propose for these cases is similar to the one suggested in the context of German VP-topicalization with ergative verbs in chapter 5 (cf. fn. 15). Suppose that the verbal default case feature ACC can optionally occur on the V-head of ergative verbs in Icelandic. Thus, when V and a DP argument together with an expletive are merged, ACC on V selects both the DP and the expletive for ACC checking since both elements are equally close. Later, finite T containing a NOM feature is merged. NOM can select an element marked ACC as its feature checker because ACC and NOM on a DP are structurally compatible (cf. also e.g. chapter and section 2.1.2 in this chapter). But let us assume then that, if a DP has already been assigned a case value, it can, at least optionally, resist assignment of a second case value. NOM can therefore be assigned to the expletive only and not to the subject DP. Finally, we may assume that NOM on T then also determines the choice of the categorial features on T, i.e. that the expletive therefore is also selected as a feature checker for the uninterpretable categorial features on T. The expletive therefore moves to T, then to the NOM checking position and finally to the clause-initial position in [Spec, CP]. Thus, we obtain an expletive-associate construction involving a DP associate. This option is restricted to ergative contexts because it is only in these contexts that two case features are available for a subject and the second case feature can attract an expletive away from a DP associate.Google Scholar
  19. 19.
    Thus, although the IO does not occur in a potential categorial checking position any more but in a case checking position (cf. step vii) and hence does not meet the general requirement for categorial attraction (cf. chapter 2, condition 21), it can nevertheless be selected as a categorial checker by D on T. The reason for this is that N on V2/3 has selected the IO as a categorial feature checker already before the IO has been moved to its case checking position and that therefore N on V2/3 could never be checked if it did not have the capacity of influencing the selection of a checker by D on T.Google Scholar
  20. 20.
    Given the observations made earlier on Korean, it should be pointed out that the situation with respect to parameter (35) may sometimes be slightly more complex than suggested in the discussion of German and Icelandic. As observed in chapter 3.2.9, Korean allows relatively free argument order. However, we saw in fn. 9 of this chapter that Korean also seems to license Dative subjects. This suggests that, as in Icelandic, an oblique case feature can be inserted in T in Korean and that therefore case features can be independent features. The question that arises then is why Korean can have free argument order and oblique subjects at the same time. The properties of Korean can be accounted for if parameter (35) is not necessarily set in a uniform way for all elements bearing case features. The features I have been focusing on in the analysis of oblique subjecthood in Icelandic are the case features on verbal heads. Let us assume then that these are indeed independent features in Korean as well. As a consequence, T does not always bear Nominative but it can also be associated with a Dative feature, and V can bear Nominative. We therefore obtain constructions with Dative subjects and Nominative objects. But suppose now that the case features on nominal elements are not independent features in Korean but subfeatures of the verbal features in the categoriel feature matrices of D and N. This means that case feature checking cannot take place before categorial feature checking has taken place. Hence, DPs only enter case checking relations above TP and we obtain a layer of case checking projections above TP where arguments can be ordered in different ways as in German. Hence, the situation in Korean suggests that parameter (35) has to be divided into two subparameters, one concerning the status of case features on verbal heads and the other one concerning the status of case on nominal heads. We then obtain the following three-way distinction for German, Korean and Icelandic: (i) case as a subfeature of both verbal and nominal heads (German); (ii) case as a subfeature of nominal heads, but as an independent feature with verbal heads (Korean); (iii) case never a subfeature (Icelandic).Google Scholar
  21. 21.
    Note that although prosodie properties may play a role sometimes with respect to argument ordering in German, stress certainly could not be considered as the crucial factor. For example there are cases of object inversion in which the 10 does not seem to bear stress at all. Consider for example: (i) Ich werde meine Fotos dem Peter erst morgen zeigen German I will my pictures (DO) the Peter (IO) only tomorrow show ‘I will show my pictures to Peter tomorrow only.’ In (i), it is the adverb modified by erst (‘only’) which is stressed and the IO does not seem to bear any stress. But cf. chapter 3, fn. 39, for a phenomenon in German which may be comparable to the Icelandic data discussed in the text.Google Scholar
  22. 22.
    Note that non-overt movement for case checking by the IO seems to have an effect not only on the distribution of the phonological features but also on the distribution of the semantically relevant properties of the 10. As pointed out above, the IX) can bind into the IO with DO-IO orders whereas the 10 cannot bind into the DO. This of course suggests that not only the phonological features of the 10 are left behind but also the semantic features, otherwise we would expect reconstruction effects. Hence, non-overt movement for case checking seems to affect only the case features and possibly the catégoriel features of the IO and the IO is therefore interpreted phonologically and semantically in the position in which it is focalized.Google Scholar
  23. 23.
    A closer look at the derivation of (42) may be necessary however. The IO has been assumed to move covertly to the DAT checking position above VP2. The question that arises then is what happens when D on finite T is reactivated for multiple attraction. The simplest analysis is the following. We may assume that the non-overt IO is attracted to TP first because it is the closer object. However, this movement is non-overt because the phonological featuresof the IO have been left behind in a position below the case checking position from where the IO gets attracted to TP. Then D on T is reactivated again and attracts the DO. In this case, there is no reason for leaving the phonological features of the argument behind and the DO therefore moves to TP overtly. Hence, we obtain the order DO-Neg-IO. It is important to note that this analysis excludes the possibility that the IO could overtly move to TP. Ungrammatical orders like the one shown in (31a) (DO-IO-Neg) therefore cannot be derived.Google Scholar
  24. 24.
    The result of this derivation is a VO order. In this respect, Icelandic has a slightly marked status from a typological point of view because, as observed in chapter 3.4, the presence of a morphological case system tends to be related to the occurrence of OV orders. The question that arises then is why OV orders generally cannot be found in Icelandic. Given the discussion in the text, one factor seems to be verb movement which undoes the effects of object leftward movement. This of course also holds for clauses involving finite main verbs since finite main verbs can be argued to generally move to C due to the symmetric V2 property of Icelandic. However, for non-finite main verbs, there may be alternative ways to derive OV order which would not be affected by verb movement. One option discussed in chapter 3.4 was to say that OV could be the result of movement of a large constituent containing the object to a proxy projection above TP. I suggested that such an approach might capture the relation between OV and morphological case if we assume that remnant movement depends on the presence of a proxy domain above TP which is independently motivated. This condition is met in a language like German. Above finite TP, the presence of a proxy domain for case checking is motivated by the presence of a case feature on T. As for non-finite T, it does not bear a case feature itself, but a case checking domain is licensed above TP because case features such as ACC or DAT on object DPs become available for checking above non-finite TP (cf. e.g. chapter 3, examples 76/77 for illustrations of case checking above non-finite TP). This is the crucial difference between German and Icelandic now. In Icelandic, object case is checked below TP due to the negative setting of parameter (35), so there is no reason to have a case checking domain above non-finite TP as in examples like (43). It could therefore be argued then that OV with non-finite verbs cannot be derived in terms of remnant movement because the proxy domain that is required for remnant movement is not available above non-finite TP in Icelandic. The consequence of such a proposal would be that the setting of parameter (35) may be a factor contributing to the absence of OV orders in Icelandic. But cf. also the discussion in section 6.1.1 below (example 51) for an additional point on the absence of OV orders in Icelandic. 25 NOM and ACC elements have the same structure and they are therefore not structurally incompatible. Hence, NOM and ACC can co-occur on the same element. Cf. chapter (passives of ditransitive verbs in German), chapter 5 (fn. 15) or fn. 18 above (ergative verbs) for contexts where this option is realized.Google Scholar
  25. 26.
    An anonymous reviewer raises the question as to what the status of CP and PP arguments is within the system proposed here for Icelandic. In chapter, I argued that, in German, case features lead to argument order variation not only with DPs but also with CPs and PPs. In Icelandic, however, word order variation with CPs and PPs generally does not seem to be attested. In general, this property is a consequence of the status of case features within the framework proposed here. Delayed case checking as proposed for German CPs/PPs in chapter is generally not possible because case features on verbal heads are not subfeatures of categorial features in Icelandic. Instead case features have to be checked immediately. However, there may be another reason why CPs/PPs do not seem to be affected by the presence of case features in Icelandic, and this factor may play a role for potential cases involving the contexts discussed in section 5.2. CPs and PPs are never morphologically marked for case in Icelandic. Furthermore, case features are independent features in Icelandicwhich are optionally present on verbal heads. It could therefore be argued that case features that have no morphological effect do simply not get generated on verbal heads in Icelandic. In German, however, case features are subfeatures of categorial features. They are therefore part of a categorial feature matrix, and the syntactic context does not have an influence on their occurrence. Case features can therefore give rise to word order variation with CPs and PPs in German but not in Icelandic.Google Scholar
  26. 27.
    At least (55b) then also involves agreement licensing in AgrsP via a non-overt expletive so that the order DO-SU is maintained. Cf. chapter 4 for the same proposal for German.Google Scholar
  27. 28.
    As for object inversion after a non-finite verb form as found in Icelandic (cf. example 39 above), there does not seem to be an equivalent in Yiddish (Molly Diesing, p.c.). Although more detailed research may be necessary to confirm this observation, it would not be surprising in terms of the proposals made here. Given that case features only can be checked after categorial feature checking, VP-internal object inversion with DPs (i.e. the order DO-IO in 56a) cannot be derived in Yiddish. The only scenario in which object inversion would be possible is with a DO NP which checks its categorial features in VP1 and which gets attracted by an ACC feature on V2 rather than V1. However, we could exclude the option of inserting an ACC feature under V2 in Yiddish by assuming that since case features are closely related to specific categorial features there is no variation with respect to where ACC is inserted with ditransitive verbs. Instead, we can assume that the case feature of the lower object is always linked to the V-feature of the lowest V-head (V1). Hence, an ACC NP can never move past an IO because ACC is checked right above VP1. Thus, the main idea would be that the variation found in Icelandic with respect to insertion of ACC on V1or on V2 is again directly related to the independence of case features and that this option is not available in other languages. In languages in which case is dependent on categorial features, case features always co-occur with the same categorial feature.Google Scholar
  28. 29.
    In example (56), the following two word orders would be grammatical, too. (i) a. Maks hot Riflcen nit gegebn dos bukh Yiddish Max has Rebecca (to) not given the book (DO) b. Maks hot dos bukh nit gegebn Rifken Max has the book (DO) not given Rebecca (lO) In (ia), the IO simply moves to the matrix domain alone. Thus, the subject and the IO both get attracted to TP for categorial feature checking and then the two arguments move to their case checking positions above TP. As for the DO it undergoes non-overt movement for case and categorial checking. For (ib), the derivation is slightly less straightforward because the DO must have moved past the IO. One possibility to account for this movement is to say that the DO is directly attracted to an ACC checking position above TP and that this process allows the DO to move past an element like the IO which bears another case feature. On its way to the ACC checking position, the DO checks its categorial features and finally the IO undergoes non-overt movement for case and categorial checking.Google Scholar
  29. 30.
    Diesing (1997:402ff.) uses the data in (55) and (56) as evidence for an A’-movement analysis of object movement in Yiddish under the assumption that reordering is only possible with A’-movement. However, given the proposals made in chapter 3 and here, it is of course not necessary to conclude that variation in argument orders can only be the result of A’-movement. The additional arguments Diesing (1997:407f.) uses for an A’-movement approach to object movement in Yiddish are the same ones as those standardly given for German or Dutch, i.e. the occurrence of PP-movement and parasitic gaps. However, in chapter 3,1 argued that both phenomena do not exclude an A-analysis of object movement. Furthermore, as Diesing (1997:407, fn. 23) points out, Yiddish also exhibits properties such as clause-boundedness of object movement which would rather suggest that this process involves A-movement.Google Scholar
  30. 31.
    Note that V1 in main clauses is productive in both languages (cf. Santorini 1994:87, Sigurðsson 1990:62). This suggests that there is an additional way to derive V1 (apart from the one outlined for subordinate clauses below) which is restricted to main clause contexts and which is also available in Yiddish. I will not pursue this issue here.Google Scholar
  31. 32.
    It should be pointed out however that, in Icelandic, the occurrence of an overt expletive in the contexts discussed here is nevertheless possible although an empty expletive would be licensed. Thus, we obtain the following optionality (example from Ottosson 1994:113) (i) a. Ég veit ekki hvort bað rignir Icelandic I know not whether it rains ‘I don’t know whether it is raining.’ b. Ég veit ekki hvort [e] rignir I know not whether rains This optionality could be argued to be related to the mixed status of C in Icelandic as a C(Agrp) head, (ia) illustrates the case where C is interpreted as a normal C-head whereas in (ib) the Agrp feature is taken into account and licenses the non-overt expletive. Note finally that Ottósson (1994:113) claims that the contrast in (i) is only available in subordinate clauses. However, the data discussed by Sigurðsson (1990:49) suggest that the same optionality also arises in main clauses, as would be expected in terms of the approach proposed here.Google Scholar
  32. 33.
    Note however that the claim here cannot be that the strategy used in Icelandic is the only one that is available to circumvent that-trace effects. Thus, for example Norwegian (cf. Engdahl 1985) and certain dialects of English (cf. Sobin 1987) also lack that-trace phenomena although these languages do not license empty expletives.Google Scholar
  33. 34.
    With some verbs, the assumption that the vowel represents number is fairly straightforward. For example, the verb nevnaa (‘to name’) has the following inflectional paradigm in the present tense (from Rohrbacher 1994:120): nevn-i (lsg), nevn-ir (2sg), nevn-ir (3sg), nevna-a (lpl), nevna-a (2pl), nevna-a (3pl). In this case,-i can be argued to correspond to singular and-a to plural. However, certain other classes of verbs have a vowel alternation in the 2nd and 3rd person singular (-ur or-ar rather than-ir, cf. Barnes and Weyhe 1994:205). This is illustrated by the verb tota (‘to throw’) (from Bobaljik 1995:46): kast-i (lsg), kast-ar (2sg), kast-ar (3sg), kast-a (lpl), kast-a (2pl), kast-a (3pl). Given the proposal made in the text, we would have to assume that the number morpheme in the present singular can be subject to allomorphy conditioned by the presence of the person morpheme-r.Google Scholar
  34. 35.
    It may be however that a clause structure where only AgNP below TP is present but no AgrpP above TP is a marked one which is prone to change. The idea would be that the default head for subject agreement is T (due to N on T) and that if only a single agreement relation is established the unmarked option is to associate agreement with N on T and hence to project an Agr projection above TP as in languages like German. Thus, the change in Faroese may lead to a tension between the morphological and the syntactic information which is available to the language learner, i.e. the morphology would favor an AgrsP above TP whereas the syntax (e.g. due the presence of oblique subjects) requires an AgrNP below TP. The plausibility of such a scenario is supported by observations made by Jonas (2002). As Jonas observes, in spoken Modern Faroese the DAT-ACC case pattern with experiencer verbs as in (63) tends to be replaced by a NOM-ACC pattern. Given the observations made before, we could explain this observation by assuming that an AgrP-CaseP-TP clause structure (i.e. a clause structure in which oblique nominal constituents cannot be subjects) indeed starts replacing a CaseP-TP-AgrP structure. Faroese may thus exhibit what Kroch (1989) refers to as competing grammars and more specifically competition with respect to the position of the agreement licensing projection within the clause structure.Google Scholar
  35. 36.
    If the proposals made in fn. 35 are on the right track (i.e. that an AgrP-CaseP-TP clause structure may start replacing a CaseP-TP-AgrP clause structure), it would not be surprising that XP-SU orders can be found in Faroese. Since Faroese has empty expletives (cf. e.g. Holmberg 1995:26, Platzack 1987:387/8), XP-SU orders could be derived on the basis of an AgrP-CaseP-TP clause structure with an empty expletive in AgrP.Google Scholar

Copyright information

© Springer Science+Business Media Dordrecht 2002

Authors and Affiliations

  • Eric Haeberli
    • 1
    • 2
  1. 1.University of ReadingUK
  2. 2.University of GenevaSwitzerland

Personalised recommendations