Concluding Summary The main objective of this study is to provide an account of the core syntactic and semantic properties of middles across a sample of languages. The starting point of the inquiry was to explore the common characteristics that pertain to middles in English and SC and identify those that hold for middles across languages. A further goal was to account for the differences between middles in English and SC; to pinpoint the domain(s) in which they occur as well as to provide the answer to the question of why they occur. The initial hypothesis was that differences between English and SC are not only relevant for these two languages, but reveal a deeper pattern of cross-linguistic parameterization. To facilitate the testing of this hypothesis, in addition to English and SC, the scope of this study includes also Dutch, French, Hebrew, Italian, and Polish. As shown in this study, Dutch, English and Hebrew pattern together as the middle formation operation applies in the lexicon in these languages. French, Italian, Polish and SC pattern together as the middle formation operation applies in the syntax (LF) in these languages. Chapters 1, 2 and 3 can be viewed as a clearing-the-grounds-for-middles unit as they provide an identity card of middles and the tools instrumental in capturing them in the languages under consideration here. Chapter 2 examines in more detail and from a critical perspective certain aspects of Reinhart s Theta System, the theoretical framework that is assumed in this study and presented in Chapter 1. Middles have previously posed some serious problems to the Theta System. For these and several other problems in the system, it was necessary to introduce some modifications. Chapter 2 focuses on the properties of feature clusters and the rules that regulate their co-occurrences. In interpretative terms, underspecified clusters are argued to be potentially problematic in that they are thematically deficient. The discussion in this Chapter sheds some light on the type of thematic deficiency of underspecified clusters and the way this thematic deficiency is repaired. By critically examining the implications of the notion underspecified clusters, I propose that a new principle the Full Interpretation of Thematic Roles (1) needs to be added to the assumptions of the Theta System. This principle is guided by the Non-Identity Constraint (2), which is a core constraint operative at the interface between the system of concepts and the computational system as well as at the interface between the computational system and semantics.
286 Concluding Summary (1) Full Interpretation of Thematic Roles For the purposes of interpretation, all clusters must be fully specified. (2) The Non-Identity Constraint An n-place verb, n>1, is encoded in terms of non-identical feature clusters. On the analysis proposed here, the Full Interpretation of Thematic Roles (FITR) and the Non-Identity Constraint ensure that all participants stand in distinct and fullydefined relation to the event in question. As an added bonus, in terms of the Full Interpretation of Thematic Roles, I provide a possible solution to one of the longstanding mysteries in linguistics the Animacy Restriction in the Double Object Construction. Further, the principle allows a better understanding of the necessity of the Cluster Distinctness Constraint as a constraint of natural language. Through the prism of the FITR, I propose a redefinition of this formal constraint (4). (4) Cluster Distinctness Revisited Two underspecified clusters are indistinct if there is a construal under which they are identical. In its empirical coverage, the Cluster Distinctness Constraint Revisited is equal to the original Cluster Distinctness Constraint. Being tied to the Non-Identity Constraint, the redefinition has explanatory power with respect to the environments this formal constraint is active in, as well as with respect to the notion of indistinct clusters. Another issue that is dealt with is the status of arguments versus adjuncts in the Theta System. As shown in Chapters 3 and 4, addressing this issue has important consequences for the account of middles in languages such as English. Recipient Goals are not allowed to realize in English middles (either as DPs or as PPs), as seen by the contrast between (5a) and (5b). Directional-goals readily realize in the middle derivation (5c). (5a) (5b) (5c) Big presents ship easily. *Big presents don t ship friends/to friends easily. Big presents don t ship easily to foreign countries. As argued in Chapter 2, the rules regulating the co-occurrences of feature clusters are tied to the Non-Identity Constraint. This facilitated the conclusion in Chapter 3 that the effects of the restriction on the co-occurrences of feature clusters that is operative in (5b), should be absent from (5c) as the rules regulating the cooccurrences of feature clusters are operative only on feature clusters of a single predicate. The issue of what should count as a feature cluster also has broader implications for the Theta System. It is argued that some adjuncts are best analyzed as not being encoded in the Theta System, whereas an alternative account was
Concluding Summary 287 proposed for others. On the account presented in Chapter 2, the thematic deficiency of the cluster of Dative Goals (i.e. [-c]) is argued to be repaired by the FITR. This naturally led to the question of the role of the Dative-to. The answer to the question led us to a fine-grained approach to prepositions. Namely, it was argued that prepositions fall into two groups: semantically vacuous prepositions and semantically contentfull prepositions. The presence of semantically vacuous prepositions was argued to be motivated by Case reasons. Following Parsons (1990), I argue that they do not appear in the logical form of a sentence. Semantically contentfull prepositions contribute to the meaning of the sentence. As such, they are always present in the logical form of the sentence. This allows us to capture the relatedness of Dative Goals and Locative Goals. Their relatedness was formally represented here by encoding them both as [-c] clusters. As argued, whereas the FITR applies in the former case, it does not apply in the latter. Another modification of the Theta System, which was introduced in Chapter 4, is establishing the existence of the underspecified ([ ]), a cluster not specified for any of the features. The feature system requires such a cluster to be well defined logically, but there has been so far no empirical evidence for its existence (which would be a serious problem). I show that this feature cluster is operative in Middle Formation in languages of the English (lexicon) type. In Chapter 3, the characteristics of middles across languages are examined. In descriptive terms, the core of what middles are can be summarized as follows: (i) (ii) Middles are characterizing - generic statements (i.e. ascribing properties to entities, involving quasi-universal reading, typically with a modal flavor of capability or potentiality). The external role of middles is not linked in the syntax, but it is always present in the semantics and interpreted as ARB with +human flavor. Given these properties, both the defining and the delimiting of the scope of the investigation into middles are facilitated. Derivations that ascribe properties to entities but do not have an implicit external role (e.g. stative or generic unaccusatives in section 2.2.2 of Chapter 3) should not be treated as middles. Eventive outputs that - for all intents and purposes - look just like middles in syntactic terms (e.g. se/si-passives in Slavic and Romance languages) should also not be confused with middles. As argued in Chapter 5, these instances are best treated as ARB-passives. Given the fact that middles are generic statements, their stativity a trait often taken to be the defining characteristic of middles in the literature - follows straightforwardly. Given the quantificational analysis of generics (cf. Krifka et al. (1995)) that I adopt here, adverbial modification often noted as a characteristic of middles can be fully understood. The analyses of Condoravdi (1989) and Krifka et al. (1995), provide, thus, the tools to formally approach our middles.
288 Concluding Summary Not all the properties of middles, however, follow straightforwardly from the presence of Gen in the logical form of middles. One needs to account for the interpretation of the implicit role as arbitrary, with +human flavor with middles in both lexicon and syntactic languages (e.g. English and SC). Further, one needs to account for the fact that the predicate of an English middle is quite peculiar as it is cannot be used to form particular sentences (cf. 2.1.3 of Chapter 3). Chapter 3 examines also the diagnostics used in the literature to tease out middle derivations. With respect to some of these diagnostics, I have argued that they are unreliable not because middles do not pass the tests in question, but because one is not sure what the tests diagnose. I have also pointed out some difficulties that arise from attempts to derive the properties of middles from the i-/s-level distinction. This discussion seems to warrant the conclusion that simply stating that middles are i-level predicates does not actually amount to much, as various researchers have shown that the contrast between s- and i-level predicates is not a uniform one, but that there are several, similar but independent, contrasts that seem to be lumped together. More importantly, the evidence presented in 2.1.2 of Chapter 3 warrants caution with respect to approaches to middles that build upon the premise that states do not have an e-role and then use this to explain why states like love cannot be felicitous input to the middle formation operation in some languages. By addressing the question of what the best way to tackle the middle is, I propose that the middle construction should be tackled cross-linguistically by means of a parameterizable operation. I have shown that the properties of the middle construction are such that in some languages it applies in the lexicon whereas in others it applies in the syntax. The conclusions reached in Chapter 3 necessitated splitting up the investigation into two separate parts: Chapter 4 deals with the middle construction in lexicon languages, where a lexicon language is a language in which middles are derived presyntactically, and Chapter 5 deals with the middle construction in syntactic languages, where a syntactic language is a language in which middles are derived in the syntax (LF). By examining the hypothesis of whether middles across-languages can be derived in the same module, I have also pointed out certain empirical and conceptual problems that arise even with highly elaborate approaches if one attempts to propose a unified analysis for middles crosslinguistically. The main aim of Chapter 4 was to present an analysis of the phenomenon of middles in the Dutch/English type of language. A brief overview of the main components of the approach is given here. Instead of a constraint like the Affectedness Constraint or the Actor-constraint, my claim here is that the Lexicon Middle Formation (LMF) should be constrained by the Lexicon Middle Formation Visibility Requirement (6). (6) The LMF Visibility Requirement A verb is visible to the LMF iff its verbal concept contains a [/+c] role.
Concluding Summary 289 This Requirement accounts for the lexical restriction present in languages such as Dutch and English. It exhaustively defines the set of verbal concepts the LMF can apply to. It defines the set of verbal concepts that disallow the LMF and captures the common denominator such verbal concepts have. The LMF deletes the content of a [/+c] cluster and creates the [ ] cluster - ARB-role. As shown in Chapter 4, the creation of this role directly accounts for the crucial properties of middles in the Dutch/English type of language: a) the presence of the implicit role and its interpretation as ARB, with +human flavor, b) the presence of Gen - standardly assumed to account for the generic reading of middles (cf. Condoravdi (1989), Krifka et al. (1995), and Steinbach (1998)) and the inability of the middle-verb to appear in particular sentences, c) the lack of morphological marking (the absence of the Case-absorbing morphology in middles in lexicon languages), d) the restriction on the co-realization of clusters of the base entry, and e) the unergativity of middles in Dutch and English. In accordance with the hypothesis that the middle formation operation is a parameterizable operation (cf. Chapters 3 and 4), the main objective of Chapter 5 was to account for the properties of middles in syntactic languages languages in which the middle formation operation applies in the syntax (LF). The first step was to show that middles in syntactic languages exhibit the cross-linguistic characteristics identified for middles in Chapter 3. This exploration allowed us to pin-point the root of the illusiveness of middles in syntactic languages. The existence of the middle comparable to the middle in Dutch and English is blurred in syntactic languages due to the fact that it shares morpho-syntax with passives, which are, semantically, quite different from middles. In both lexicon and syntactic languages, middles are generic statements. Both types of middles obligatorily retain the implicit external role, which is interpreted as ARB, with +human flavor. In both types of languages, the operation that derives middles is ARB-Genericization. Since the middle-formation operation is bound to different modules, lexicon and syntactic languages use different tools to reach their common semantic goal. Whereas lexicon languages utilize the ARB-role i.e. the [ ] cluster, syntactic languages utilize ARB-saturation. Namely, in lexicon languages, the process is the arbitrarization of a cluster, while in syntactic languages, it is the arbitrarization of a variable. This, in turn, is perfectly consistent with the levels at which these two types of arbitrarization apply. Whereas arbitrarization in lexicon languages applies in the lexicon, where there are only clusters, arbitrarization in syntactic languages applies at LF. As argued further in this Chapter, ARB-saturation of Chierchia (1995a) is not just an invaluable tool in accounting for impersonals in a variety of syntactic languages, but it extends to middles and ARB-passives in such languages as well as to the Object-Arbitrarization construction, which is pervasive across Slavic languages. As noted, without appealing to this special type of saturation, it is quite difficult to account for the fact that in all three constructions, the implicit role is interpreted as an unspecified, generally plural, human entity. The representation of English (7a) is given in (8a) and of its SC middle counterpart (7b) in (8b).
290 Concluding Summary (7a) (7b) (8a) (8b) Tristram Shandy reads easily. Tristram Šendi se lako čita. Gen e, x arb [reading (e) & [-c-m] (e, Tristram Shandy) & [ ] (e, x arb )] [easy (e, x arb )] Gen e, x arb [reading (e) & [-c-m] (e, Tristram Shandy) & [+c+m] (e, x arb )] [easy (e, x arb )] As argued throughout this study, the differences between the lexicon and syntactic languages follow from the fact that the middle formation operation is a parameterizable operation. Whereas some of the differences follow from the very fact that middles are derived at different levels (e.g. the availability of ECM-middles in syntactic and the unavailability of ECM-middles in lexicon languages), others follow from the way the operation is executed in these two distinct modules (e.g. restrictions on the co-realization of clusters, which is active in the lexicon but not in syntactic languages). As shown on the sample of languages in this study, the choice of the parameter-setting does not seem to be determined for each of the parameterizable operations individually. Rather, it follows from the generalization in (9) and holds for various parameterizable operations with respect to a single language. (9) The Lexicon-Syntax Parameter (Reinhart and Siloni (2003)) UG allows thematic arity operations to apply in the lexicon or in the syntax (LF). This, in turn, makes the setting of the parameter during acquisition much easier: by setting the parameter for one parameterizable operation; the child fixes the parameter for all of them. As further argued in Chapter 5, the existence of the ARBpassives in syntactic languages is not just explained - it is actually predicted on the account here. In lexicon languages, the presence of the underspecified [ ] cluster enforces the Gen operation. This cluster is derived by cluster manipulation, which is permitted only in the lexicon and cannot take place in the syntax. Since this cluster is not present in syntax languages, there is nothing to enforce Gen. Among the intriguing properties of middles in syntactic languages under consideration here is the fact that they are marked with the presence of the clitic SE. As noted in Chapter 1, a further puzzling characteristic is that the same clitic marks a variety of outputs, which, at the first glace, do seem to have much in common. The attempt to explore this issue and the unwillingness to accept a mere coincidence as the explanation of the appearance of the same marker across a variety of derivations, led to the exploration in Chapters 5 and 6 of the underlying generalization pertaining to the presence of this clitic. The assumption here is that the unifying approach to clitic SE can be achieved if the clitic is viewed as Case-absorbing morphology, as proposed in Reinhart and Siloni (2003). Evidence from various languages that utilize
Concluding Summary 291 the clitic SE is consistent with its role as a case-absorber. Furthermore, the arguments in Chapter 6 corroborate the hypothesis that the clitic is not associated with a theta-role. Such a unifying approach to the clitic is both conceptually and empirically desirable. I have provided evidence that the governing generalization behind the SE-constructions under consideration here cannot be found in terms of either their shared semantics or syntax, or in terms of their historical relatedness. The governing generalization behind the various SE-constructions seems to be that they are outputs derived by the application of one or another arity operation. Middles continue to present themselves as an important field of research in both empirical and theoretical terms. Though the main objectives of this study have been met, the exploration of the phenomena dealt with in this study is not exhausted. Some of the issues that are left for further research have been mentioned throughout this study. Just like in many other research projects on middles (cf. Keyser and Roeper (1984), Zubizarreta (1987), Cinque (1988), Fagan (1992), Hoekstra & Roberts (1993), and Ackema and Schoorlemmer (1994) to name just a few), the middle construction has also been used here as a testing ground for the interface between the system of concepts, the computational system (i.e. syntax), inference (i.e. semantics), and discourse. A particular issue that I would like to note again here is the relevance of middles for the on-going lexicalist versus non-lexicalist dispute. In this respect, it is of paramount importance to stress that, as shown on a sample of languages in this study, the differences between middles across-languages can be straightforwardly captured only if one works under the hypothesis that the language faculty contains an active lexicon. Namely, as shown in Chapters 4 and 5, a whole range of properties that lexicon and syntactic middles exhibit follows from the fact that Middle Formation is a parameterizable operation. As argued here, the parameterization, in turn, can only be accommodated in a model that contains an active, dynamic component of the lexicon. The complexity of the phenomenon of middles resulted in an interconnectedness among Chapters that encompasses more than one issue. As already pointed out, Chapters 4 and 5 can be viewed as a continuation of the discussion regarding the structure of the lexicon, which is opened in Chapter 1. Chapter 1 presents Siloni s (2003) arguments pertaining to the domain of reflexivization across different languages, which can be captured only if one assumes that in certain languages reflexivization is bound to the lexicon, whereas it applies in the syntax in others. Chapter 4 deals with middles in English and Dutch, the proper treatment of which presupposes the existence of a manipulative rule driven - component of the lexicon, whereas in Chapter 5, the validity of the Lexicon-Syntax Parameter is tested against the data in Hebrew. The behavior of middles in lexicon languages prompted an investigation into the domain of arguments and adjuncts (Chapters 2, 3 and 4). Middles in syntactic languages, on the other hand, necessitated a closer look at the broader spectrum of SE-constructions (Chapters 5 and 6). Last but not least, I hope that the study as a whole has shown how enormously intriguing the issues on the interface between the System of Concepts and the CS, on the one hand, and the interface between the system of concepts and semantics, on the other hand, are.