To serve double duty under syncretism - How Nanosyntax and grafting account for the free relative construction

To serve double duty under syncretism

How Nanosyntax and grafting account for the free relative construction

1

Introduction

A well-known observation from the literature (cf. Ingria 1990) is that conflicting features in syntax may exist as long as the realization of those features is identical. For example, in (1a) we see that under normal conditions conflicting features lead to an ungrammatical result (1a):

(1) a. *John or I buy candy. b. John or Mary buys candy.

In (1b) we see that the DP John and the DP Mary require the same syntactic feature and hence (1b) is grammatical. A third situation arises when the features are conflicting as in (1a), but the situation is resolved by the fact that the same morphophonological form is used to spell out both these features. Such a case is known from so-called free relatives in German:

(2) a. Wer who.nom nicht not stark strong ist, is(nom) muss must(nom) klug smart sein. be ‘Whoever is not strong, must be smart.’

b. *Wer who.nom Gott God schwach weak geschaffen shaped hat, has(acc) muss must(nom) klug smart sein. be ‘Whoever God has created weak must be clever.’

c. *Wen who.acc Gott God schwach weak geschaffen shaped hat, has(acc) muss must(nom) klug smart sein. be

‘Whomever God has created weak must be clever.’ (German, Groos and Van Riemsdijk 1981, p. 177) d. Was what.nom/acc du you mir me gegeben given(acc) hast, have ist is(nom) prächtig. wonderful

‘What you have given to me is wonderful.’ (German, Groos and Van Riemsdijk 1981, p. 212)

In each sentence in (2), a single form has to satisfy two case feature requirements. First, the relative pronoun has to satisfy the requirement of the predicate or verb in the main clause, and, second, it has to satisfy the requirement of the predicate or verb in the relative clause. In (2a), both predicates (ist stark ‘is strong’ andmuss klug sein ‘must be smart’) require nominative case, which is satisfied by the nominative relative pronounwer ‘who.nom’. In this sentence, the case requirements of both clauses match and the sentence is grammatical.

Syntactic case requirements are conflicting in (2b) and (2c). The verbgeschaffen ‘shaped’ requires accusative case and the predicatemuss klug sein ‘must be smart’ requires nominative case. The nominative relative pronounwer ‘who.nom’ in (2b) is only able to satisfy the nominative case requirement. In (2c), the accusative relative pronounwen ‘who.acc’ only satisfies the accusative case requirements. The case requirements of the clauses are conflicting, different morphophonological forms are used to spell out the features and the sentences are ungrammatical.

(2d) shows an example in which the features are conflicting, situation is resolved by the fact that the same morphophonological form is used to spell out both these features The verbgegeben ‘given’ requires

accusative case andist prächtig ‘is wonderful’ requires nominative case. Because was ‘what.nom/acc’ is used as both the realization of the nominative and the accusative neuter relative pronoun, the sentence is fine. Apparently, in free relative constructions, multiple case feature requirements can be satisfied simultaneously under one of the two following circumstances.1Consider the following schematic notions of grammatical derivations, in which X and Y are morphosyntactic features which are realized as the morphophonological forms P and Q. (3) a. X→ P Y→ Q X = Y b. X→ P Y→ Q X̸= Y P = Q

In (3a), the morphosyntactic feature X is identical to the morphosyntactic feature Y, e.g. they are both nominative case. An example of such a derivation is shown in (2a). In (3b), the morphosyntactic features X and Y are not equal, but the morphophonological forms P and Q are identical. An example of such a derivation is shown in (2d).

This phenomenon poses three challenges for linguistic theory. First, somehow the syntax has to allow multiple conflicting features on a single syntactic node, while still being able to rule out (1a). For instance, if a relative pronoun functions simultaneously as a subject and an object (as in example (2d)), it has to bear both nominative and accusative case features. How can we construct a syntax that permits such a situation while it forbids the situation in (1a)? Second, there should be a single morphophonological form available that corresponds to these multiple conflicting features. So, in (2d), a single lexical entry must bear simultaneously nominative and accusative cases feature and can therefore be the subject in the main clause and the object in the relative clause. In other words,was ‘what.nom/acc’ serves double duty. Third, we face the challenge to build a syntax that allows for such double duties. Put differently, we need to explain why the grammar has, apart from a sentence as (4), also the option to realize the same meaning (more

1

I would like to stress here that satisfying multiple case requirements happens simultaneously. In that respect, it should be distinguished from ambiguity. Also in cases of ambiguity one may say that a particular form has two functions (two ‘meanings’) , but not at the same time, as shown in (i) (Ingria 1990, p. 194).

(i) a. The sheep is ready.

b. The sheep are there.

c. *The sheep that is ready are there.

In (ia),sheep has a singular meaning, and (ib) refers to the plural. (ic) shows that this singular and plural meaning cannot be combined in one sentence, i.e.sheep can only be either singular or plural and not both at the same time (Ingria 1990, p. 194). This differs fromwas ‘what’ in (2d), which is simultaneously both nominative and accusative.

Whether something is singular or plural refers directly to the real world: a sheep cannot be both a single sheep and two sheep at the same time. For nominative and accusative this semantic reference is absent, as they only refer to the syntactic positions in the sentence and not to differences in meaning (Ingria 1990, p. 194; Sauerland 1995). It is not the syntax, but rather the semantics that lead to a crash of the derivation.

economically) as (2d). (4) Was what.nom/acc du you mir me gegeben given(acc) hast, have das that.nom/acc ist is(nom) prächtig. wonderful

‘What you have given to me (that) is wonderful.’ (German, Groos and Van Riemsdijk 1981, p. 212)

In this thesis, I will propose an analysis within Nanosyntax (Starke 2009) to meet the first two challenges. In Nanosyntax, each feature corresponds to its own terminal node. Caha (2009) has proposed a universal hierarchy for case, in which more complex cases are built from less complex cases, e.g. the accusative is built from the nominative. I will claim that the solution to the first challenge follows naturally from Nanosyntax: as cases are structurally built from each other, case-features are never conflicting. The solution to the second challenge follows from phrasal spellout and the superset principle, which form the basis for the spell-out operation in Nanosyntax. I will show that syncretic cases are spelled out by the lexical entry of the biggest case involved (i.e. the case containing the other case). As for the third challenge, I will use a type of merge called grafting, as described by Van Riemsdijk (2006b). Grafting is a type of merge in which an embedded node is merged again with a different structure. The result is that two clauses share a particular node, the so-called ‘callus’. I will show that a sentence is grammatical when the shared features correspond to the same lexical entry in both clauses.

This thesis is structured as follows. In section 2, I will show how the free relative construction poses three challenges, of which two are specifically challenging for theories of late insertion, such as DM. In section 3, I will discuss the free relative construction in detail. I will conclude that syncretic relative pronouns and syncretic prepositional phrases can serve double duty, and that a relative pronoun can satisfy case requirements from within a prepositional phrase. Section 4 describes, on the one hand, how previous analyses in a Lexicalist framework account for serving double duty under syncretism, and, on the other hand, why Lexicalism should be abandoned for other reasons. In section 5, the theory of Nanosyntax is introduced as the theory that I will adopt. In section 6, which is the core of the morphological analysis, I will a give Nanosyntactic analysis of why a form can serve double duty under syncretism. Section 7 elaborates on possible syntactic analyses of free relative constructions. I will conclude that grafting accounts best for placing two conflicting features on a single syntactic node. In section 8, which is the core of the syntactic analysis, I discuss how grafting and Nanosyntax complement each other. In section 9, I discuss phenomena to which the analysis could be extended.

2

Three challenges of free relative pronouns

In the previous section, I showed that free relative pronouns can serve double duty when they are syncretic between two cases. I will explain in this section why this poses three challenges, of which two are specifically challenging for theories of late insertion, such as DM.

As I have shown in the German example in (2d) the phonological forms of the nominative and accusative neuter relative pronoun are the same, i.e. the forms are syncretic. This is not the case for (2b) and (2c): the masculine relative pronoun has separate forms for nominative and accusative case. So, comparing (2d) to (2b) and (2c), one could conclude that the phonological form (i.e. whether the forms are syncretic) influences the grammaticality of the derivation. To be more precise, when a form is syncretic between two cases, it can simultaneously satisfy case requirements of both cases. These facts pose three challenges, of which two are especially challenging for theories of late spellout.

First, the syntax has to allow multiple conflicting features in a single syntactic position. As we have seen above, this is far from trivial since data such as the ones in (1) show that we cannot simply give up the requirement that features should be matching. So, the question is why this requirement is strictly obeyed in (1) where it seems to be violated in (3). The answer I propose is that indeed this is only seemingly the case; in actual fact, the features match. The step that we need to take in order to fully understand this is offered by nanosyntax in which features are structurally stacked on top of each other. As a consequence the nominative case is contained in the accusative case, rather than that it contains a different or conflicting case-feature. Second, a lexical entry should be allowed to correspond to multiple conflicting features. This is not particularly new; in many theories of morphology among which DM, lexical entries are underspecified for their featural make-up.2As such it is unsurprising that a single lexical entry may function as the spell-out of different case-features. Lastly, a specific syntactic structure is required that allows relative pronouns to serve double duty. I will explain these challenges in further detail, but in order to appreciate these challenges more fully we first sketch the currently widely accepted - albeit with many minor theoretical differences - theory of Distributed Morphology.

In DM, there is no lexicon as in earlier Lexicalist theories, but the functions of the lexicon in this type of approach are distributed among other components of the grammar. In DM, the syntactic hierarchical structure goes all the way down: elements within both syntax and morphology enter the same type of constituent structure as can be captured in binary branching trees. I adopt the figure that sketches the grammar of DM below from Harley and Noyer (1999, p. 3).

2

DM generally uses the term vocabulary item for what is called lexical entry in Nanosyntax. To avoid confusion, I will adopt the term lexical entry throughout the paper for the items in both theories.

morphosyntactic features

syntactic operations

morphological operations logical form

phonological form lexical entries conceptual interface encyclopedia list 1 list 2 list 3

Figure 1: The Y model of grammar in DM

The syntax generates structures by combining morphosyntactic features via the operations merge and move (Harley and Noyer 1999). In this stage, case assignment takes place. Note here that only features (and roots) enter the derivation and not words with a phonological form as is the case in a classic Lexicalist approach (Embick and Noyer 2007). These features are merged into a particular structure that is sent to PF and LF for phonological and logical (semantic) interpretation respectively. LF exhibits meaning-related structural relations, such as quantifier scope. Finally, (part of ) the structure may match with a structure that is stored in the so-called encyclopedia which contains a list of idioms of the language. The encyclopedia relates lexical entries to meanings, taking the entire derivation of the expression into account (Embick and Noyer 2007). At PF, morphological operations, such as fusion and fission, are first applied, which are relevant for word formation. The result is a structure with terminal nodes containing bundles of features (Harley and Noyer 1999). At that point, lexical entries are inserted into terminal nodes in a process called spellout, in which these lexical entries compete for insertion. The item that is selected is the most specific one which still contains a subset of the features present in the syntax (subset principle, Halle 1997). This means that lexical entries do not need to be fully specified for the syntactic positions where they are inserted. When no more specific lexical entry is available, the default is inserted (Embick and Noyer 2007). Let us now turn back to why satisfaction of multiple case requirements of free relative pronouns is problematic in this theory. First, I will argue why it is problematic that the syntax has to allow for multiple conflicting features in a single syntactic node. In DM, case assignment is a syntactic operation. During these syntactic operations, the two conflicting features are copied into a single syntactic position. Normally, syntax does not allow two conflicting features in a single syntactic position (cf. Sauerland 1995, p. 8), as is shown in (5).

(5) a. Mary or John rides. b. Mary or they *rides/*ride.

In (5a), bothMary and John both require verb agreement in the third person singular, and the sentence is grammatical. However, in (5b),Mary requires third person singular but they requires third person plural agreement. In this sentence, the plural and singular feature cannot coincide in a single syntactic position and the derivation crashes. In contrast, in (2d), the feature conflict is allowed (under specific conditions). Consequently, an adaption of DM is needed to allow for feature conflicts in syntax.

Within DM (in cf. Asarina 2011) it is proposed that the theory can be adapted in such a way that it allows for feature conflicts in the syntax. It is argued that the conflict can be resolved later if a single lexical entry is inserted into the node which corresponds to both the conflicting features. This is the second challenge I will discuss, two which the answer does not seem to follow immediately from DM.(2d) requires a lexical entry which is selected as nominative and as accusative. However, the lexical entry cannot be specified for both cases. In (2d), both nominative and accusative are required, butwas ‘what.nom/acc’ is also used as either nominative or accusative in sentences. According to the subset principle, the item that wins the vocabulary insertion competition is the most specific one which still contains a subset of the features. In (6), I give the three required lexical entries.

(6) a. [nom, acc]↔ /was/ b. [nom]↔ /was/ c. [acc]↔ /was/

For a sentence such as (2d), (6a) is needed, but for a sentence in whichwas ‘what.nom/acc’ either nominative or accusative case requirements, (6b) or (6c) is required. A solution would be to argue all three lexical entries are required. However, as DM is developed to account for the mismatch between phonology and syntax, this is not a desired solution.

Another possible solution is to argue thatwas ‘what.nom/acc’ is selected as the default. A problem arises then only if there are two case syncretisms in a paradigm. An example of a language in which this problem arises is Polish, which besides a syncretic nominative and accusative, also has a syncretic genitive and dative which are different from the nominative and accusative (Dyta 1984, p. 703). I give two lexical entries for the default in (7).

(7) a. [ ]↔ która ‘who.fem.nom/acc’ b. [ ]↔ której ‘who.fem.gen/dat’

(7a) should be inserted as nominative and accusative and (7b) should be inserted as genitive and dative. However, as both rules are the default, they cannot be distinguished. One of the two entries has to be specified for a case to be able to keep them apart. In sum, we need a lexical entry that corresponds to multiple conflicting features which can also be inserted if only one of the features is required.

Besides the two challenges DM faces for free relative constructions, the construction is a third issue that is not specifically problematic for DM but that poses a more general problem for any run-of-the-mill

syntactic theory. That is, a specific syntactic structure is required that allows relative pronouns to serve double duty. It is not the case that all syncretisms can simultaneously satisfy different requirements. Below I show an example in which the form of an adjective can have either nominative or accusative case. The example shows that it cannot bear the two cases simultaneously.

(8) *den the.acc glücklichen happy.acc/gen Mannes man.gen

‘the happy man’ (German, Sauerland 1995, p. 6)

This means that the free relative construction has such a syntactic structure that the free relative pronoun is allowed to serve double duty that it can simultaneously have two different functions. In this thesis, I will investigate which syntactic conditions are required to enable a single syncretic form to satisfy multiple case requirements.

In sum, free relative constructions pose three challenges. First, the syntax has to allow multiple conflicting features in a single syntactic position. Second, a lexical entry should be allowed to correspond to multiple conflicting features. Lastly, a specific syntactic structure is required that allows relative pronouns to serve double duty.

3

Free relative construction

In the previous section, I introduced a phenomenon in which a form can serve double duty under syncretism for which I will propose an analysis in this thesis. In this section, I provide a detailed description of this phenomenon: the free relative construction. I will first introduce free relative pronouns and give examples of them in several languages. I show that, in general, so-called matching effects are required, i.e. the category of the phrase (determiner phrase or prepositional phrase) or the case of the relative pronoun should be identical in both the relative and the main clause. Exceptions are made, and non-matching is allowed, when the relative pronoun or preposition is syncretic between two categories or cases. Finally, I will also show that a relative pronoun can satisfy case requirements from within a prepositional phrase.

3.1 Terminology

Relative clauses are used to modify part of the main clause, typically a determiner phrase. A relative clause is introduced by a relative pronoun, which refers to its antecedent: the modified phrase. In regular relative clauses, the antecedent of the relative pronoun is overt. Both the antecedent and the relative pronoun have their own syntactic position in the main and the relative clause (Van Riemsdijk 2006a). An example from Dutch with the relative pronoundat ‘that’ is shown in (9).

(9) Piet Piet ziet sees het the huis house dat that Jan John gebouwd built heeft. has ‘Piet sees the house that John built.’

The relative clause introduced bydat modifies the overt antecedent huis ‘house’. Het huis ‘the house’ is the object ofziet ‘sees’ in the main clause and dat ‘that’ is the object of gebouwd heeft ‘has built’ in the relative clause.

In contrast, in a free relative clause, the relative pronoun has no overt antecedent in the main clause. Rather, the free relative pronoun refers to the free relative clause itself. The relative pronoun takes a syntactic position in the relative clause and the whole relative clause fills a syntactic position in the main clause (Van Riemsdijk 2006a). An example is shown in (10).

(10) Piet Piet ziet sees wat what Jan John gebouwd built heeft. has ‘Piet sees what John built.’

In (10),wat ‘what’ has no antecedent in the main clause. Instead, it refers to the relative clause wat Jan gebouwd heeft ‘what John built’ as a whole. Wat Jan gebouwd heeft ‘what John built’ functions as the object ofziet ‘sees’ in the main clause and wat ‘what’ is the object of gebouwd heeft ‘has built’ in the relative clause.

Bresnan and Grimshaw (1978) noticed that free relatives show so-called matching effects, i.e. free relatives are only acceptable when the category of the phrase or the case that is required in both clauses is matching. The categories that they refer to are determiner phrases or prepositional phrases. The examples

below from Groos and Van Riemsdijk (1981, p. 173) show categorial matching. (11) a. *Ken know jij you met with wie who zij she flirt? flirts

‘Do you know with who she is flirting?’ b. Ken know jij you wie who zij she net just kuste? kissed ‘Do you know who she just kissed?’

In (11), the verbken ‘know’ and kuste ‘kissed’ require a determiner phrase and the verb flirt ‘flirt’ requires a prepositional phrase introduced by the prepositionmet ‘with’. Sentence (11a) is ungrammatical as there is a single position available in the sentence for two non-matching categories: a determiner phrase forken ‘know’ and a prepositional phrase forflirt ‘flirt’. (11b) is grammatical as both ken ‘know’ and kuste ‘kissed’ require a determiner phrase (Groos and Van Riemsdijk 1981, p. 173).

In languages with morphological case, not only the category of the phrase has to be matching, but the specific case as well. I repeat the examples from (2) in the introduction here, which show that sentences are only grammatical when the same case is required for both verbs (Groos and Van Riemsdijk 1981, p. 177).

(12) a. Wer who.nom nicht not stark strong ist, is(nom) muss must(nom) klug smart sein. be ‘Whoever isn’t strong must be clever.’

b. *Wer who.nom Gott God schwach weak geschaffen shaped hat, has(acc) muss must(nom) klug smart sein. be ‘Whoever God has created weak must be clever.’

c. *Wen who.acc Gott God schwach weak geschaffen shaped hat, has(acc) muss must(nom) klug smart sein. be ‘Whomever God has created weak must be clever.’

Example (12a) is grammatical because bothist stark ‘is strong’ and muss klug sein ‘must be smart’ require a nominative. Examples (12b) and (12c) are ungrammatical, sincemuss klug sein ‘must be smart’ requires a nominative andgeschaffen hat ‘shaped has’ requires an accusative. To sum up, we may conclude that free relatives are only grammatical if the category of the phrase or case requirements in both clauses converge.

However, the sentence in (13) (repeated from (2d)) indicates, surprisingly, that under certain circum-stances non-matching is allowed. It seems that when a relative pronoun is syncretic between two cases, it is able to satisfy two case requirements simultaneously (Groos and Van Riemsdijk 1981, p. 212).

(13) Was what.nom/acc du you mir me gegeben given(acc) hast, have ist is(nom) prächtig. wonderful ‘What you have given to me is wonderful.’

In (13),gegeben ‘given’ requires accusative case and ist prächtig ‘is wonderful’ requires nominative case. The neuter relative pronoun is syncretic between the nominative and the accusative, i.e. the nominative as well as the accusative iswas ‘what’. Apparently, although the predicate and the verb have different case requirement, (13) is grammatical, because in both cases the formwas ‘what’ is required.

We can conclude that a free relative pronoun can serve double duty and bear conflicting case values. One condition for this is that the surface form of each individual item is identical. This suggests that formal feature conflicts can be resolved by syncretism. In other words, serving double duty is allowed under syncretism.

The next section gives examples of free relative pronouns that serve double duty under syncretism in other languages with morphological case.

3.2 Free relatives with morphological case

As we have seen in German, the neuter free relative pronoun is syncretic between nominative and accusative in (13). In Russian, it is the masculine relative pronoun has the syncretic formkogo ‘who.acc/gen’ for both the accusative and the genitive (Levy and Pollard 2002, p. 222).

(14) a. Kogo who.acc/gen ja I iskal, sought(acc) ne not bylo was(gen) doma. home ‘Who I was looking for wasn’t at home.’

b. *Cego what.gen ja I iska, sought(acc) ne not bylo was(gen) doma. home ‘What I was looking for wasn’t at home.’ c. *Cto what.acc ja I iskal, sought(acc) ne not bylo was(gen) doma. home ‘What I was looking for wasn’t at home.’

Izvorski (1997, p. 279) gives another example from conversational Bulgarian, where nominative relative pronouns can be used instead of the accusative, resulting in a syncretism between the nominative and the accusative. (15) a. Šte will celuna kiss(acc) kojto who.nom/acc dojde comes(nom) prâv. first ‘I will kiss whoever comes first.’

b. *Šte will celuna kiss(acc) kogoto who.acc dojde comes(nom) prâv. first ‘I will kiss whoever comes first.’

(15a) is grammatical, because the formkojto ‘who.nom/acc’ is syncretic between nominative and accusative case and it is able to satisfy the case requirements in both clauses. (15b) is ungrammatical, becausekogoto ‘who.nom’ is only able to satisfy the nominative case requirement ofceluna ‘kiss’ and not the nominative case requirement ofdojde ‘comes’.

The examples from Polish and conversational Bulgarian have shown that, just as the German example in (13), non-matching of cases is allowed under syncretism. The main and the relative clause have different case requirements. However, since the relative pronouns have syncretic forms for the required cases, the sentences are grammatical.

case.

3.3 Free relatives with prepositional case

In several languages (e.g. Dutch, Spanish, English) some cases are not expressed morphologically but by means of a prepositions and a relative pronoun. In Dutch,wie ‘who’ is always used when the referent is a person. For the nominative (subject) and accusative (object) only the relative pronoun is used, i.e. there is a single syncretic form for both cases.3 As I illustrated in the previous sections, the relative pronoun can simultaneously satisfy different case requirements in the main and relative clause.

(16) Hij he bewondert admires(acc) wie who(nom/acc) de the wedstrijd game wint. wins(nom) ’He admires who wins the game.’

In (16),wie ‘who.nom/acc’ satisfies simultaneously the nominative and accusative case requirements of bewondert ‘admires’ and wint ‘wins’. This is no longer surprising, since we now know that a relative pronoun can satisfy multiple case requirements at the same time. Next, I will show that the combination of a preposition and a relative pronoun (i.e. a prepositional phrase) can also satisfy more than one requirement for the category of the phrase. Bresnan and Grimshaw (1978) use the term category to distinguish a determiner phrase from a prepositional phrase. First I will show that a prepositional phrase can satisfy multiple requirements twice as the prepositional phrase. Then I will illustrate that a prepositional phrase can also satisfy multiple requirements once as the prepositional phrase and once as the determiner phrase which is contained in the prepositional phrase.

I have already shown that a relative pronoun can satisfy different case requirement in a relative and main clause. Next, I will show that a prepositional phrase can also serve double duty and satisfy different requirements for the category of the phrase. The restrictions on satisfying multiple requirements of the category for the phrase are similar to the restrictions on case requirement. The prepositional phrase has to be of the same category (prepositional phrase or determiner phrase).

(17) a. Ik I wacht wait op on wie who ik I verliefd in love ben. am ‘I wait for whom I am in love with.’ b. *Ik I speel play met with wie whom het the huis house is. is

Intended: ‘I play with the one whose house it is.’ c. *Ik I speel play van of wie who het the huis house is. is

Intended: ‘I play with the one whose house it is.’ d. *Ik I zit sit op on wie who ik I verliefd in love ben. am ‘I sit on whom I’m in love with’.

3

In certain word orders the indirect object (dative) can also be used without preposition. I leave this out of the discussion, because it does not change anything about the point I am trying to make.

In (17a),op wie ‘for/with whom’ can be regarded as similar to wer ‘who.nom’ in (12a), because they satisfy both requirements of the verb withop wie ‘for/with whom’. Since both verbs in (17b) and (17c) require different prepositions, the sentences are ungrammatical. These sentences can be compared tower ‘who.nom’ in (12b) andwen ‘who.acc’ in (12c), because they all have a different phonological form. Lastly, in (17d), both verbs require the prepositionop ‘on’, but the sentence is ungrammatical. This follows from the fact that op wie ‘on/with whom’ is used with different meanings: it has to be simultaneously an abstract preposition and a literal location.

An example similar to (17a) from another language comes from Spanish (Grosu 2003, p. 29).

(18) Soñaba I-dreamed con with quien who tú you saliste went-out ayer. yesterday

‘Lit. I dreamed about with whom you went out yesterday.’

In (18),con quien ‘with/about you’ satisfies the requirement of soñaba ‘I dreamed’ and the requirement of thesaliste ‘went out’ (Grosu 2003, p. 30).

Apparently, both a relative pronoun bearing morphological case and a prepositional phrase can satisfy two different case requirements or requirements for the category of the phrase in a single sentence on the condition that they are syncretic.

The third and last observation is that the relative pronoun contained in the prepositional phrase can also satisfy a second requirement on its own. (19) shows how the relative pronounwie ‘who’, contained in the prepositional phraseop wie ‘for who’, also satisfies case requirements within the relative clause. (19) a. Ik I wacht wait op on wie who wint. wins(nom) ‘I wait for the one who wins.’ b. Ik I wacht wacht op on wie who jij you kust. kiss(acc) ‘I wait for whom you kiss.’

In (19), the verbwacht ‘wait’ requires the prepositional phrase op wie ‘for who’. Besides being part of the prepositional phrase,wie ‘who.nom/acc’ also satisfies on its own the nominative case that the verb requires in the relative clause in (19a). In (19b), it additionally satisfies the accusative case required bykust ‘kiss’. So, this seems to indicate that the relative pronoun is still available to satisfy case requirements from within a complex structure.

Van Riemsdijk (2006a, p. 343) claims that the order is also important in these examples. He argues that examples such as (19b) are grammatical, but similar constructions are ungrammatical when the main and relative clause are reversed (Van Riemsdijk 2006a, p. 343), as in (20) (or in (11a)).

(20) *Jij you kust kiss op on wie who ik I wacht. wait ‘You kiss for whom I wait.’

the relative clause. The prepositionop ‘on’ in the main clause selects a determiner phrase object, which is the first constituent of the relative clause. In such cases, there is nothing which causes the sentence to be ungrammatical. In (20), however, the relative clause has a prepositional phrase and follows the main clause. The verb in the main clause,kust ‘kiss’, requires a determiner phrase, but a preposition intervenes, causing the ungrammaticality of the sentence.4

In the previous sections, I have shown that relative pronouns obligatory satisfy the case requirements and requirements for the category of the phrase in both the main and the relative clause. There is one exception where non-matching is allowed, i.e. when the two (prepositional phrases including the) relative pronouns are syncretic between two cases or categories. I have also shown that a relative pronoun is available to satisfy case requirements from within a prepositional phrase that also satisfies its own requirements. Throughout the rest of this thesis, I will focus on these two observations.

4

I do not entirely agree with this judgment. To me, the sentence in (20) is somewhat marked, but far from ungrammatical. The sentence could be used, for instance, when the speaker wants to emphasize the event of kissing.

4

Previous accounts in Lexicalism

In the previous section I have shown that the phonological form seems to influence the grammaticality of the sentence. This suggests that we need a theory in which the phonological form is available to the syntax. Such accounts have been proposed within Lexicalist frameworks (cf. Zaenen and Karttunen 1984). First, I will illustrate why underspecification in Lexicalism provides solutions to two of the challenges I formulated in the introduction: multiple conflicting features are allowed on a syntactic node and a single lexical item can correspond to multiple conflicting features. I discuss these accounts in this thesis, as they make use of underspecification, as I will in my analysis. However, an important difference is that the underspecification proposed in Lexicalism is based on a stipulation: it is proposed solely to account for the free relatives. I will show how the underspecification is independently motivated within Nanosyntax. Moreover, I will refer to arguments against Lexicalist approaches in general.

In this section, I will discuss several analyses of free relative constructions within Lexicalism, each of them involving underspecification. Before I go into the details of the analysis, I will briefly introduce Lexicalism.

Most theories of grammar assume that particular items are drawn from a lexicon for syntactic compo-sition. In traditional Lexicalist approaches (cf. Chomsky 1970, Halle 1973), the items that are fed to the lexicon are words. Morphology and syntax are held completely distinct. I adopted the figure below from Marantz (1997, p. 202). meaning syntax sound sound lexical combining meaning

Figure 2: Lexicalist view of computational lexicon

The Lexicalist hypothesis comes in two versions. In the weak version (cf. Siegel 1974), derivation is part of the lexicon (lexical combining in figure 2) and inflection is part of syntax. In the strong version (cf. Chomsky 1970), both derivation and inflection take place in during lexical combining. Words enter the syntax with appropriate syntactic features, with a phonological form and with a meaning. After syntax, the syntactic structure is interpreted phonologically and semantically.

Zaenen and Karttunen (1984) do not discuss free relative constructions specifically, but related phenom-ena. To my knowledge, they were one of the first to propose an underspecification analysis for ‘serving double duty’ constructions. Zaenen and Karttunen (1984) discuss whether homonyms should have a single form in the lexicon or several separate forms each with their own feature specifications. The example they give is the verbtake. One item for take would be specified for: ‘verb, present tense or no tense, no third person singular’. Two examples of the several fully specified items would be: ‘verb, present tense, first person singular’ and ‘verb, present tense, second person plural’. They argue for the first alternative, proving their point with an example from Icelandic in which a noun bears simultaneously accusative and dative case (Zaenen and Karttunen 1984, p. 312).

(21) Hann he stal stole(dat) köku cookie.acc/dat og and borðaði ate(acc) köku. cookie.acc/dat ‘He stole and ate a cookie.’

Apparently, the representation that people have in their lexicon forkokü ‘a cookie’ is: ‘noun, accusative and dative case’. Besides representations in the lexicon, Zaenen and Karttunen discussed which features cannot be neutralized. One example they give is shown below (Zaenen and Karttunen 1984, p. 315).

(22) *Ich I habe have den the Dozenten teacher(s).acc.sg/dat.pl gesehen seen(acc) und and geholfen. helped(dat) ‘I have seen the teacher and helped the teachers.’

In (22),den Dozenten ‘the teacher(s)’ should be a singular object of gesehen ‘seen’ and a plural object of geholfen ‘helped’, as it is syncretic between the two. However, den Dozenten ‘the teacher(s)’ cannot be used to refer simultaneously to a singular and a plural object. Zaenen and Karttunen (1984, p. 316) therefore formulated the following constraint:

(23) Anti-pun ordinance: a phrase cannot be used in two different senses at the same time.

In other words, the ungrammaticality of (22) is a result of a semantic clash rather than a syntactic clash.5 Pullum and Zwicky (1986, p. 751), cal the phenomenon “phonological resolution of syntactic feature conflict”. According to them, it is the phonology that influences grammaticality directly Pullum and Zwicky (1986, p. 751): “the fact that the conflicting morphosyntactic requirements lead accidentally to similar phonological consequences is sufficient to prevent the conflict from causing ungrammaticality”. They formulated a principle that captures their generalization (Pullum and Zwicky 1986, p. 766).

(24) The Resolution Principle (RP): A syntactic feature conflict on a factor in coordination can be resolved if (a) particular values of the features are syntactically imposed on the factor; and (b) a phonological form is available which is, at the relevant [lexical?] stratum of representation, ambiguous between these values.

Part (a) is similar to what Zaenen and Karttunen called the Anti-pun ordinance: the features should be purely syntactic and not semantic. In part (b), Pullum and Zwicky describe syncretic form as ambiguous representations.

Ingria (1990) applied this type of reasoning specifically to the free relative construction under discussion. I repeat the examples from (2b) and (2d) below.

(25) a. *Wer who.nom Gott God schwach weak geschaffen shaped hat, has(acc) muss must(nom) klug smart sein. be

‘Whoever God has created weak must be clever.’ (Groos and Van Riemsdijk 1981, p. 177) b. Was what.nom/acc du you mir me gegeben given(acc) hast, have ist is(nom) prächtig. wonderful 5

‘What you have given to me is wonderful.’ (Groos and Van Riemsdijk 1981, p. 212)

Ingria argues that purely formal features can be neutralized producing truly underspecified representations Ingria (1990, p. 200).

(26) a. wer↔ (nom) b. wen↔ (acc)

c. was↔ (: or (nom (acc)) d. ist, war↔ (nom)

The agreement of formal features involves a check for non-distinctness. The verbgeschaffen ‘shaped’ in (25a) requires accusative case, which is non-distinct from the accusative case inwen ‘who.acc’. However, muss ‘must’ requires nominative, which is distinct from the accusative case in wen. The grammaticality of (25b) follows as the sentences require nominative and accusative andwas ‘what.nom/acc’ non-distinct from both cases.

Also in the head-driven phrase structure grammar (HPSG) framework researchers gave analyses in terms of underspecification. For example, Sag (2003, p. 274) proposed: “...feature structures will be specified for all features declared appropriate for their type but the values of those features need not be assigned leaf (maximal) types in the type hierarchy.” A leaf maximal type refers to a single case feature, a non-maximal type contains two case features, for instance nominative and accusative. So, in other words, feature structures need to be specified for case, but not necessarily for only a single case.

Each of the analyses mentioned here can account for the described data. However, the feature specifi-cation seems to be rather stipulative, as in (26). Even though an account in Lexicalism seems to run into lesser difficulties than an analysis DM, Lexicalism should be abandoned for other reasons.

In the 90s, Lexicalism and the special status the Lexicalists attributed to words was challenged by DM (Halle and Marantz 1993). Words were said to have a special sound, special meaning or a special structure/meaning correspondence, which turned out not be true (cf. Marantz 1997). It is argued that words can be built from morphemes via merger just as constituents are built from words via merger. I will adopt this view and not provide an analysis within Lexicalism.

In the introduction I already illustrated why DM is not able to account for the phenomenon of serving double duty under syncretism. In this section I have shown that in a Lexicalist approach it is possible to formulate a description for why free relative pronouns serve double duty. However, these proposals are purely stipulative and Lexicalism runs into several difficulties as a theory too. What I need is a theory of late spellout, which has a direct mapping between syntax and phonology, which uses underspecification and that has a direct connection between different case features. In the next section, I will introduce the framework of Nanosyntax as the theory I will adopt.

5

Nanosyntax

Serving double duty under syncretism can be explained best within Nanosyntax, which I will introduce in this section. This theory will be the basis for a solution to two of the challenges I formulated: the syntax has to allow for conflicting features on a syntactic node and a single lexical entry has to be inserted to correspond to conflicting features. In this section, I will provide some background information about Nanosyntax which is necessary to understand the details of analysis. I will start by giving a broad overview of the theory and discuss how it differs from DM. Next, I will elaborate on lexical entries, i.e. mapping rules between syntax and phonology. Subsequently, I will discuss the case hierarchy as Caha (2009) proposed it, in which more complex cases always structurally contain less complex cases. Lastly, I will show briefly how case assignment is handled within Nanosyntax.

5.1 How Nanosyntax differs from DM

Nanosyntax is, just as DM, a theory of late insertion, i.e. phonological forms inserted into syntactic slots that are generated by the syntax. DM and Nanosyntax differ primarily in that the latter allows for phrasal spellout: whereas in DM only endnodes are open to insertion, in Nanosyntax it is allowed to spell out higher nodes and therefore entire syntactic phrases. In general, here is a mismatch between the number of features and the number of phonological form, i.e. a single affix can correspond to a number of features (e.g. -s in English corresponds to third person, singular, present tense). To account for this mismatch, DM has to

use certain morphological operations. In this section, I will show that if we allow for phrasal spellout, the need for this postsyntactic morphological component disappears, i.e. morphological operations such as fusion and fission are no longer needed.

Starke (2009) developed the framework of Nanosyntax, of which the Y model of grammar looks as follows as in figure3. atomic features merge ‘big tree’ SMS lexicon PF LF

Figure 3: Starke’s version of the Y model of grammar

Syntax combines atomic meaningful features by merge into a big tree. This tree is not only responsible for syntax and morphology ( just as in DM), but also for formal semantics, which is normally not considered to be part of syntax. Starke calls this module SMS: Syntax-Morphology-Semantics. The structure that leaves the SMS module is handed over to the lexicon and, via lexical access, it is interpreted phonologically and

conceptually Caha (2009, pp. 51-52). In the model of the grammar in figure 3 the morphological component that is part of the model in DM is absent. According to Starke, this component is no longer necessary when spellout of non-terminals (also ‘phrasal spellout’) is allowed. In the remainder of this section I will discuss how phrasal spellout replaces the morphological operations fusion and fission.6

Caha (2009, pp. 57-60) argues that phrasal spellout is equivalent to terminal spellout and fusion. The implication is that it is better to have one operation (spellout) than two operations (spellout and fusion). He illustrates his point using suppletive negative forms in Korean, as discussed by Chung (2007). Sentences in Korean can be negated by one of the two negative prefixesani or mos, as shown in (27).

(27) a. ca sleep -n -pres -ta -decl ‘is sleeping’ b. mos/an(i) neg ca sleep -n -pres -ta -decl

‘cannot sleep / is not sleeping’ (Korean, Chung 2007, p. 97)

The verbal- ‘know’ does not combine with one of these markers, but has the suppletive form molu-. (28) a. al know -n -pres -ta -decl ‘know(s)’ b. *mos/*an(i) neg al know -n -pres -ta -decl c. molu neg.know -n -pres -ta -decl

‘cannot know / does not know’ (Korean, Chung 2007, p. 115)

The verbal- ‘know’ can be causativized by -li, which results into a verb meaning ‘to inform’. With this complex verb, the non-suppletive form of the negation is used and the suppletive form is ungrammatical.

(29) a. ani neg / / mos neg al know -li -caus -ess -past -ta -decl ‘did not/could not inform’

b. *molu neg.know -li -caus -ess -past -ta -decl

‘did not/could not inform’ (Korean, Chung 2007, p. 120)

Chung (2007) argues that only when the negation is a sister ofal- ‘know’ suppletion occurs, as in (30a). The suppletive form does not appear in the causativized structure becauseal- ‘know’ is first causativized and only later negated, as in (30b) (Chung 2007, pp. 130,132).

6

(30) a. NegP V know Neg b. NegP vP v caus V know Neg

In Nanosyntax, the lexical entry for the negatormolu is the structure in (30a). In DM, an operation called fusion is required, which changes the structure into a node containing two features. This forms the structural part of the lexical entry.

(31) /molu/↔ [Neg, know]

Caha (2009) notes that fusion occurs on two conditions: sisterhood (the negation and the verb ‘know’ should be sisters to avoid the causative form to be suppletive) and content of the node (al- ‘know’ is suppletive, ca-‘sleep’ is not). These two conditions lead to a paradox. Fusion (an operation that happens before lexical insertion) should only happen if the insertion provides a suppletive form (the result of lexical insertion). So, in order to know whether operation X should happen, information about the outcome of an operation Y which follows the operation X is required.

Since Nanosyntax allows for spellout of non-terminals, both pieces of information (sisterhood and content of the node) are available in the lexical entry itself. For the other form, the terminal nodes can be spelled out one by one. In sum, phrasal spellout achieves the same result as fusion and spellout of terminal nodes. Actually, it also avoids the paradox mentioned before and the separate component morphology does not have to be assumed Caha (2009).

Besides fusion, the morphological operation fission is also used in DM. Fission applies to a bundle of features and splits them into distinct terminals, which each correspond to a lexical entry. Fission is used for, for instance, agreement features, which are assumed to be under a single terminal but are spelled out by distinct forms (Harley and Noyer 1999). In Nanosyntax this operation is not required since each feature has its own terminal node, so at no point in the derivation features are bundled.

In conclusion, if phrasal spellout is adopted, the architecture of the grammar as we know it from DM can be simplified. To be more specific, the need for certain morphological operations (e.g. fusion, fission) that occur in the separate morphology component of grammar disappears.

5.2 Lexical entries

As I mentioned in the previous section, phonological forms are mapped on syntactic features in Nanosyntax. Lexical entries are constructed as mapping rules which take the syntactic structure as their input and produce phonological representations as their output ( just as lexical entries in DM). To be more precise, a

phonological form can correspond to either a single terminal node or, what is different from DM, an entire syntactic phrase (i.e. phrasal spell-out).

The competition for insertion in Nanosyntax is led by two conditions: the superset principle and the elsewhere condition.

(32) Superset principle: A phonological exponent is inserted into a node if its lexical entry has a (sub-)constituent that is identical to the node. (Starke 2009, p. 3)

(33) Elsewhere condition: In case two rules, R1 and R2, can apply in an environment E, R1 takes precedence over R2 if it applies in a proper subset of environments compared to R2. (Starke 2009, p. 4)

Let me give two abstract lexical entries.7 (34) A⇔ XP YP [Z] [Y] [X] (35) B⇔ YP [Z] [Y]

In a language with only (34) as a lexical entry,A is inserted if the available syntactic structure is XYZ, but also if it YZ or Z (superset principle) are provided by the syntax. In a language with both (34) and (35) as lexical entries,A is inserted if the syntactic structure is XYZ. B is inserted if the syntactic structure is YZ (elswhere condition), andB is inserted if the syntactic structure is Z (elsewhere condition and superset principle).

In Nanosyntax, syncretisms are handled as having a single lexical entry. So, if (34) is only available in a language,A is the syncretic form for X, Y and Z. An example from natural language is the comitative and instrumental in German are syncretic and realized by the prepositionmit ‘with’.

(36) /mit/⇔ Comitative Instrumental

[E] [F]

The lexical entrymit is selected as the comitative, because it maps exactly on it. The instrumental is also realized asmit, because the instrumental is contained in the structure (superset principle) and there is no more specific lexical entry (elsewhere condition).

7

Throughout this thesis I put letters that encode features between square brackets to avoid confusion with heads (e.g. A as adjective and D as determiner).

5.3 Cases in Nanosyntax

As we have indicated above, one of the challenges that the matching under syncretism phenomenon poses is that somehow the syntax has to allow for ‘clashing’ features. In my view this can be solved by making use of the case-hierarchy proposed in Caha (2009). Using the framework of Nanosyntax, Caha (2009) proposes a containment hypothesis for case in which cases are built from each other. The structure corresponding to Caha’s case hierarchy is shown in (37).

(37) Comitative Instrumental Dative Genitive Accusative Nominative DP [A] [B] [C] [D] [E] [F]

In the tree, a nominative DP is a type of constituent in which the DP is the complement of the feature [A]. The accusative DP is built by addition of the feature [B]. In order words, cases are complex forms: cases higher in the hierarchy always contain cases lower in the hierarchy (i.e. the accusative always contains the nominative, the genitive always contains the nominative and accusative, etc.).

Caha (2009) gives some empirical arguments for this claim. First, the realizations of the different cases show formal relations, i.e. syncretisms are contingent in the hierarchy (Caha 2009, pp. 7,36). Second, the structure in (37) correctly predicts an implicational hierarchy (Caha 2009, p. 49): i.e. languages expressing a certain case will also express cases lower in the hierarchy, but not necessarily the other way around.

Cinque (2005) has shown that it is a language specific property to which structural position a noun can move in an extended projection with different nominal modifiers (adjective, numeral and demonstrative). Below I show to which position the noun moves in respectively Dutch and Farsi.

(38) a. die those drie three hele very dikke thick boeken books ‘those three very thick books’ b. un those seta three ketabe books kheili very koloft thick

‘those three very thick books’ (Caha 2009, p. 28)

In Dutch, the noun remains in base position, so it does not move. In Farsi, it appears between the numeral and the adjective, so it has moved above the adjective. In other languages a noun moves even higher. In Maasai, the noun moves above the numeral and appears between the demonstrative and the numeral, and, in Kiitharaka, the noun moves above the demonstrative, appearing before the demonstrative (Caha 2009, p. 28).

Caha (2009) argues that we also find variation in how high a noun can move in the region where case features are generated. Similar to the domain of nominal modifiers, it is language specific where the NP moves in a structure of case features in this extended projection (Caha 2009, pp. 27-30). Positions higher than the highest landing site are expressed with prepositions and are combined with the most complex case possible (Caha 2009, p. 43), positions lower than the highest landing site are expressed with morphological case.

Caha shows that in Bulgarian the highest landing site for a noun is above the accusative: Bulgarian expresses nominative and accusative case morphologically and the cases higher with functional prepositions. Genitive case is expressed by the combination of a functional preposition and accusative morphological case:na men-e ‘of me-gen’ (Caha 2009, p. 77). German is somewhat more complicated in the sense that it realizes morphological case on the determiner rather than the noun itself. German has four morphological cases: nominative, accusative, genitive and dative. In other words, the NP in German cannot be moved higher than the dative and all higher cases are expressed by a preposition, which always combine with the dative case. For instance, the comitative is expressed by a preposition combined with the dative case:mit dem Hund ‘with the.dat dog’ (Caha 2009, p. 65).

Caha (2009) uses the peeling theory, as proposed by Starke, for case assignment. Starke argues that nouns are base-generated with a number of case projections on top of them. In each movement the case phrase strands (at least) one of its case shells (i.e. the KP is ‘peeled’ under movement). Starke adopts the insight from Chomsky (1995)’s checking theory that a particular case emerges as a consequence of attraction. The structure contains positions which need particular cases in their specifier position, e.g. S-Acc (S as k-selector) requires an accusative.

Furthermore, there is a distinction between the relation verb phrases and functional prepositions have with the NPs they select. For functional preposition, the relation is non-relational: the movement of the NP within the case hierarchy does not establish any new relation between the features that are spelled out as a preposition and those spelled out as morphological case. There is a language-specific inherent relationship with the case the preposition selects. For instance, the German instrumental always combines with the dative case, which is in German the highest possible case in the hierarchy. The English genitive preposition of k-selects accusative case, as the NP can only move above the accusative (Caha 2009, pp. 40-41). For verbs, on the other hand, a new relation is established and the relation is called relational: the case is determined by a relation between a noun phrase and a k-selector. These relations are not language-specific (i.e. it does not concern how high a noun can move in the extended projection), but it is universal, e.g. finite clauses always require nominative case.

I will illustrate this with an example and a structure in (39).

b. S-Com Com Ins Dat Gen Acc Nom [A] [B] DP [C] [D] [E] [F] vP S-Nom Nom DP [A] I← with← him← dance←

The verb phrase of whichdance is the head has a k-selector for nominative (for its subject) and a k-selector for comitative (for its complement), so it takes the constituents which are required by the k-selectors. The comitative has in English a non-inherent relationship with the accusative, as the accusative is the highest possible case.

6

Current Nanosyntactic analysis

In this section I want to illustrate why serving double duty under syncretism can be explained best within Nanosyntax. I solve the first two challenges I formulated in the introduction: (1) the syntax has to allow multiple conflicting features on a syntactic node and (2) a single lexical entry is inserted that corresponds to multiple conflicting features.

First, I will discuss how Nanosyntax avoids the problem of multiple conflicting features on a syntactic node. In Nanosyntax, each feature corresponds to its own terminal node, so at no point in the derivation there is more than one feature on a node. Caha (2009) proposed a case hierarchy, in which cases are complex forms that are structurally built from each other. In this presumably universal hierarchy more complex cases contain less complex cases, but they are not conflicting, in the sense that one case is positively specified for a feature that is negatively specified in another case. Rather than a cross-classification of cases, Caha proposes a hierarchical classification.

Second, I will elaborate on how a single lexical entry can correspond to multiple conflicting features. As mentioned above, case features are part of a case hierarchy and phrasal spellout is allowed in Nanosyntax. Syncretic cases are spelled out by the lexical entry of the biggest case (i.e. the case containing the other case). The smaller case is inserted via the superset principle. When two forms are syncretic and contingent in a hierarchy, they correspond to a single lexical entry. So, syncretisms are underspecified: multiple forms correspond to a single lexical entry. Note here that the underspecification is not stipulative as it is in Lexicalist approaches, but that Caha’s case hierarchy is independently motivated.

In conclusion, I will show that a situation in which multiple (possibly different) cases are assigned to the same DP, results in a grammatical sentence if both clauses require the same lexical entry. First, I will show how the analysis for morphological case, and, second, I will give a similar analysis regarding prepositional case.

6.1 Relative pronouns with morphological case

I repeat the examples under discussion from (2) and (2d) below for convenience.

(40) a. Wer who.nom nicht not stark strong ist, is(nom) muss must(nom) klug smart sein. be ‘Whoever isn’t strong must be clever.’

b. *Wer who.nom Gott God schwach weak geschaffen shaped hat, has(acc) muss must(nom) klug smart sein. be ‘Whoever God has created weak must be clever.’

c. *Wen who.acc Gott God schwach weak geschaffen shaped hat, has(acc) muss must(nom) klug smart sein. be

‘Whomever God has created weak must be clever.’ (Groos and Van Riemsdijk 1981, p. 177) d. Was what.nom/acc du you mir me gegeben given(acc) hast, have ist is(nom) prächtig. wonderful

Each of the relative pronouns in the sentences in (40) requires two case features, either twice the nominative (as in (40a)) or once the nominative and once the accusative (as in (40b)-(40d)). Although they seem fairly similar, note that (40b) and (40c) differ from (40d) in that the latter is grammatical whereas the former two are not. The key here is that there is a single lexical entry for both the nominative and accusative neuter relative pronoun (was ‘what.nom/acc’) and there are two separate entries for the masculine (wer ‘who.nom’ andwen ‘who.acc’). I will illustrate this in the remainder of this section. For each sentence, I will start by giving the syntactic structure. Then, I will show which language specific movements take place before the features are spelled out. Finally, I will show the lexical entries that spell out the case (and gender) features. I will illustrate that multiple case-assignment is fine if the same lexical entry is inserted in both the main and the relative clause.

Let me start by giving the analysis for (40a). A relative pronoun consists of the following features: the definite determiner containing an elliptical (empty) NP (Wiltschko 1998, p. 179) (to which I will refer as a complex DP in the structures below), gender features and case features. In line with the above assumptions, the morphosyntactic features for the relative pronoun in the main clause of (40a) are the ones shown in (41), and the morphosyntactic features for the relative pronoun in the relative clause are the same.8 (41) Nom Gender DP ... masc [A]

In order to be able to express a suffix, the DP moves above the nominative, as shown in (42).

(42) Nom Gender DP ... masc [A] DP

The lexical entries that correspond to (42) are shown below.

(43) a. /w-/⇔ DP ... b. /-er/⇔ Nom GenderP masc [A] 8

In sum, in (40a),ist stark ‘is strong’ requires the nominative case in the main clause, and muss klug sein ‘must be strong’ requires the nominative case as well. So, for both sentences, the entries in (43a) and (43b) are selected, which results in a grammatical sentence.

Next, we turn to the examples (40b) and (40c). In line with the above assumptions, the morphosyntactic features for the relative pronoun in (40b) and (40c) for the main clause is (41) and for the relative clause is (44). (44) Acc Nom Gender DP ... masc [A] [B]

Here we see that the challenge of multiple conflicting features on a syntactic node is solved. As each feature corresponds to its own terminal node, there are no conflicting features on a node. To continue, I first spell out the main clause. Just as in (40a), first the DP moves above the nominative (as in (42)) and the features are spelled out by the entries in (43).

The morphosyntactic features corresponding to the relative pronoun in the relative clause are spelled out. If these features correspond to the same lexical entry, it is predicted that the structure is grammatical. However, they do not correspond to the same lexical entry and the structure is not grammatical, because German has a suppletive form for the accusative:wen. The syntactic movement before spellout for the accusative relative pronoun looks as follows.

(45) Acc Nom Gender DP ... masc [A] [B] DP ...

The lexical entries that correspond to this structure are (43a) and the one shown below. Since the accusative always contains the nominative (Caha 2009), the entry looks as follows.

(46) /-en/⇔ Acc Nom GenderP masc [A] [B]

So, in (40b) and (40c),muss klug sein ‘must be smart’ requires again the nominative case, so (43b) is selected. However,geschaffen ‘shaped’ requires the accusative case and (46) is selected. So, the two clauses require two different lexical entries for a single syntactic position. There is only a single syntactic position available, in which a single lexical entry can be inserted. (40b) and (40c) require two different lexical entries, so the derivation crashes.

Next, I will discuss an example in which the second challenge is solved: a single lexical entry is inserted that corresponds to multiple conflicting features. Nanosyntax distinguishes sentences as (40d) from sentences as (40b) and (40c). The masculine relative pronoun has a separate lexical entry for the nominative (-er in wer ‘who.nom’, (43b)) and a lexical entry for the accusative (-en in wen ‘who.acc’, (46)). The neuter relative pronoun in German, however, is syncretic between the nominative and accusative and there is a single lexical entry: (49). Again, I can make specific use of the containment hypothesis in Nanosyntax: the accusative always contains the nominative.

In line with the above assumptions, the morphosyntactic features for the relative pronoun in the main clause of (40d) are the ones shown in (47), and the morphosyntactic features for the relative pronoun in the relative clause are the shown in (47b).

(47) a. Acc Nom Gender DP ... neut [A] [B] b. Nom Gender DP ... neut [A]

As before, I spell out the morphosyntactic features corresponding to the relative pronoun in both clauses. The syntactic movements before spellout look as follows for the respectively the accusative and the nominative.

(48) a. Acc Nom Gender DP ... neut [A] [B] DP ... b. Nom Gender DP ... neut [A] DP ...

It is predicted that, if the features correspond to the same lexical entry, the structure is grammatical. This is exactly what happens. In the relative clause, the lexical entries (49) and (43a) map exactly on the morphosyntactic features in (48a). In the main clause, because the nominative case is contained in-as ‘what.nom/acc’ ((49), superset principle) and there is no more specific lexical entry (elsewhere condition), (47b) selects (49) as well. (49) /-as/⇔ Acc Nom GenderP neut [A] [B]

So, the single lexical entry in (49) gives us simultaneously two forms: the nominative and accusative neuter relative pronoun. There is only a single syntactic position available for a relative pronoun, so the lexical entries that both sentences require should be identical. That is the case and the sentence is grammatical.

Similar underspecified lexical entries can be made for the syncretic free relatives in other languages. (50a) shows the lexical entry for (14a) in Russian, (50b) the one for (15a) in colloqial Bulgarian and (50c) the one for (16) in Dutch.9

9

I assume here that in Russian and Bulgarian the DP element of the relative pronoun is spelled out as ak- and the DP element in Dutch it is spelled out as aw-, just as in German.

(50) a. /-ogo/⇔ Gen Acc Nom GenderP com [A] [B] [C] b. /-ojto/⇔ Acc Nom GenderP com [A] [B] c. /-ie/⇔ Acc Nom GenderP com [A] [B]

In Russian, there is a lexical entry for the nominative common relative pronoun (which I do not show here) and a lexical entry for the syncretic accusative and genitive, as shown in (50a). The relative pronoun can satisfy simultaneously accusative and genitive case requirements. Colloquial Bulgarian has a lexical entry for the nominative and accusative common relative pronoun, as in (50b). Just as for the German neuter relative pronoun, it can satisfy nominative and accusative case requirements. The lexical entry in (50c) for Dutch also satisfies simultaneously nominative and accusative case requirements.

In this section I have proposed a solution to the first two challenges as formulated in the introduction. First, the problem that the syntax has to allow multiple conflicting features on a syntactic node is avoided. Case features are in a structural containment relation, so features are never conflicting. Second, there should be a single form available that corresponds to these multiple conflicting features. I have shown that syncretic forms are spelled out via phrasal spellout with an underspecified lexical entry. These syncretic forms are able to satisfy both case feature requirements for which they are specified. When two different case features are required by two clauses for a single position, the sentence is only grammatical if the forms bearing the features are syncretic, i.e. they correspond to a single lexical entry (as in (40a) and (40d)). If the forms are not syncretic and do not correspond to a single lexical entry, the sentence is ungrammatical (as in (40b) and (40c)).

In the next section I will discuss how the first two challenges can also be solved when prepositional case is used. I will show why a whole prepositional phrase can serve double duty and why a relative pronoun can also satisfy its own case requirements from within the prepositional phrase. Just as for morphological case, I will illustrate that serving double duty of the same constituent is only possible when the requirements from both clauses converge on the same lexical entry, even if the exact feature specifications in the separate clauses may differ.

6.2 Relative pronouns with prepositional case

I repeat the Dutch examples under discussion from (17a), (17d), (19b) and (19a) below for convenience.

(51) a. Ik I wacht wait op on wie who ik I verliefd in love ben. am ‘I wait for who I’m in love with.’ b. *Ik I zit sit op on wie who ik I verliefd in love ben. am ‘I sit on whom I’m in love with’. c. Ik I wacht wait op on wie who.acc jij you kust. kiss ‘I wait for who you kiss.’ d. Ik I wacht wait op on wie who.acc wint. wins ‘I wait for who wins.’

As before, I will start by giving the syntactic structure for each sentence. Afterwards, I will elaborate on the language specific movements taking place before the features are spelled out. Finally, I will discuss which lexical entries spell out the relevant features.

Let me start with the sentence in (51a). Before I get to the analysis, some words about the required case are in order. Bothwacht ‘wait’ and verliefd ben ‘in love am’ require prepositional case, which is in Dutch expressed by a combination of the prepositionop ‘on’ and the accusative case. I assume here that op ‘on’ expresses an abstract prepositional (or locative) case. As prepositional case is not present in every language, I did not mention it in the universal case hierarchy as proposed by Caha. According to Caha (2009, pp. 121,130), when (abstract) prepositional case appears in a language, it is situated between genitive and dative case in the hierarchy. The part of the case hierarchy I will adopt here is shown in (52).

(52) Prepositional Genitive Accusative Nominative [A] [B] [C] [D]

In line with the above assumptions, the morphosyntactic features for the relative pronoun in the main clause of (51a) are the ones shown in (53), and the morphosyntactic features for the relative pronoun in the relative clause are the same.10

10

(53) Prep Gen Acc Nom Gender DP ... com [A] [B] [C] [D]

It is a language-specific property of Dutch that the DP can only move as high as the accusative case (i.e. all cases higher in the hierarchy are spelled out by prepositions). I show this movement below.

(54) Prep Gen Acc Nom Gender DP ... com [A] [B] DP ... [C] [D]

The lexical entries that correspond to (54) are shown below.

(55) a. /w-/⇔ DP ... b. /-ie/⇔ Acc Nom GenderP com [A] [B] c. /op/⇔ Prep Gen [C] [D]

In sum, bothwacht ‘wait’ from the main clause and verliefd ben ‘in love am’ from the relative clause require prepositional case, which is expressed by the same lexical entry. As both clauses require the same lexical

entry, the sentence is grammatical.

Next, I study what seems to be an instance of an accidental syncretism. Again, before I get to the analysis, some discussion is in order about the required preposition. In (51a), the prepositionop ‘on’ was used as an abstract prepositional. In (51b), however,op also indicates an actual location. The forms of the prepositions are identical, but they seem to correspond to a different position in the case hierarchy.

Caha (2009, pp. 128-129) argues that there are actually three different types of ‘locational’ cases, on which only two are touched upon here. I already mentioned the abstract prepositional case, which is located between the genitive and the dative. Now I will introduce the locative which has a position between the dative and the instrumental. I show the revised case hierarchy below.

(56) Locative Dative Prepositional Genitive Accusative Nominative [A] [B] [C] [D] [E] [F]

According to Caha (2009, p. 128) the prepositional is mostly used when the object is a container or a surface, or as abstract preposition. In (51a), ‘op’on is used as an abstract preposition. The object of the locative is oriented with respect to a spatial axis, either up/down or front/back (Caha 2009, p. 129). In (51b), the object is on (top of ) somebody, so the up/down axis is used.

The dative in Dutch is expressed by several prepositions (e.g. aan ‘to’ and voor ‘for’), but not by op ‘on’. In other words, there is no lexical entry in Dutch that is syncretic between prepositional, dative and locative. Instead, the locativeop ‘on’ has a lexical entry different from (55c), which is shown below. (57) /op/⇔ Loc Dat Prep Gen [C] [D] [E] [F]

Now we arrive at the spellout of the derivation in (51b). In line with the above assumptions, the morphosyn-tactic features for the relative pronoun in (51b) required by the main clause is (58) and the ones required by the relative clause is (53).

(58) Loc Dat Prep Gen Acc Nom Gender DP ... com [A] [B] [C] [D] [E] [F]

Here we see that the answer we gave above to the first challenge is also applicable here: there are no conflicting features that appear on a single node, but the more complex case (locative) contains the less complex one (prepositional). The movement before spellout for the complement ofverliefd ben ‘in love am’ in the relative clause is shown in (54). The movement of the complement ofzit ‘sit’ in the main clause is shown below. (59) Loc Dat Prep Gen Acc Nom Gender DP ... com [A] [B] [C] [D] [E] [F] DP ...

As just mentioned, the two cases do not correspond to a single lexical entry. So, this implies that this construction does not fulfill the requirement that the same lexical entry should be used to serve double duty; hence the structure is ill-formed. The lexical entries that are used in the relative clause are (55). In the main clause, however, the entry (55c) is substituted by (57).

Turning our attention to (51c), I will now answer why a relative pronoun can simultaneously be part of a prepositional phrase and satisfy case requirements as a DP. I have shown that the verbverliefd ben ‘am in love’ requires the prepositional case, which in Dutch consists of a prepositionop ‘on’ which always