An Incorporation Analysis Revisited

AN INCORPORATION ANALYSIS REVISITED Mireia Llink Universitat Autbnoma de Barcelona The syntactic behaviour of a number of constructions in Catalan involving an auxiliary and either a participle or an infinitive indicates that the two elements are very tightly knit, they behave as a unit. Following Baker's (1988) incorporation proposal, this mechanism was proposed for the Catalan verbal sequences that display a unit-like behaviour (Llinb (1990)). However, Kayne's (1993) thwry, where phrase markers determine linear ordering, may seem to imply that movement of the sort posited for the Catalan verbal sequences is mled out. More specifically, Kayne's theory imposes restrictions on the direction of adjunction. In this article we briefly review the incorporation proposal and reconsider it taking into account Kayne (1993). We conclude that despite the fact that the type of phrase marker posited for these verbal sequences does not determine the order of the terminal symbols it contains, it is a permitted phrase marker because it canies information of nodes below the zero level. O. Introduction The aim of this article is to re-examine the explanation of the syntactic behaviour of two elements which have been analysed as a result of the process of incorporation (Llink (1990)) in the light of an apparently problematic prediction made by Kayne (1993). If incorporation is regarded as the adjunction of a head to another head, certain serious problems arise which should make us reconsider the adequacy of such an analysis for these sequences. Nevertheless, if we take Robert's (1991) proposal of making a distinction between types of incorporation, the analysis need not be invalidated. In section 1 we briefly review the incorporation analysis; in section 2 we look at the sequences in more detail following Roberts (1991); in section 3 we Catalan Working Papers in Linguistics (CWPL) 3.1: 139-152 Universitat Autbnoma de Barcelona consider Kayne's (1993) proposal and his predictions for right adjunction and we review the structure proposed for the sequences in the light of Kayne's theory; in section 4 we consider alternative analyses for these sequences; and, in section 5, we conclude that Kayne's theory should be maintained for zero projections, but not for subhead structure. 1. The Sequences as the Result of Incorporation Catalan verbal sequences of the kind illustrated in (1) and (2) behave like a unit in many respects. (1) La seva filla ha aprovat I'examen. the her daughter has passed the-exam 'Her daughter has passed the exam.' (2) La seva filla va aprovar I'examen. the her daughter PAST pass the-exam 'Her daughter passed the exam.' The sequences in (1) and (2), from now on halva sequencesl, do not allow interruption -(3)- nor the application of any movement processes -(4), and (5). (3) a. *La seva filla ha sempre aprovat els exhens. the her daughter has always passed the exams 'Her daughter always passed the exams.' b. *La seva filla va no aprovar l'examen. the her daughter PAST not pass the-exarn 'Her daughter did not pass the exam.' Note, though, that this behaviour is also true 01 epistemic modal sequences, for ex. La Maria deu saber moltes coses 'Maria must know many things'. See Llinh (1991). (4) a. *Aprovat I'examen sí que ha la seva filla. passed the-exam yes that has the her daughter 'Her daughter does have passed the exam.' b. *Aprovar I'examen no va la seva filla, per6 aprendre molt si. pass the-exam not PAST the her daughter, but learn much yes 'Her daughter did not pass the exam, but she did learn a lot.' (5) a. *Ha la seva filla aprovat I'examen? has the her daughter passed the-exam 'Has her daughter passed the exam?' b. *Va la seva filla aprovar l'examen? PAST the her daughter pass the-exam 'Did her daughter pass the exam?' Given the close relation between the two verbal elements, the proposed analysis was the incoporation of the participle - V2onto the auxiliary -V 1 -, as in the following structure:2 (6) is a simplified version of the more adequate structure: (i) AGR P I AGR ' AGR-S -?Ns P I Incorporation of the participle to the auxiliary explains the cohesion of the two elements; once they have incorporated they behave like a head in almost all respects, as (3), (4) and (5) show? The head behaviour of two originally independent heads is explained by a syntactic process that does not lead to the total morphological amalgamation of the two elemenk4 Note that if the Steps 1 and 2 have been omitted because they are not relevant to the process examined here; i.e., the right adjunction of V2 to VI. V2+Agr-O moves up to V1 as a unit, anyway. Steps 3 and 4 have been omitted again because they are not directly relevant to the issue of the article and, moreover, Vl+V2 move up to higher nodes together, as a unit. Examples (5a.b) show that V1 alone cannot move up to other nodes. The issue of whether V1 is a functional category or a lexical category is not relevant to the essential idea explained here. We leave this issue open. Also, note that an objection to a structure like (i) on the basis of its violating the Structure Preserving Principle implies a misconception of the nature of functional nodes Agr and Tns. These contain the same type of information as VI, as they trigger incorporation obligatorily (see Roberts (1991)) and, therefore, must be claimed to be x-~. In certain dialects of Catalan there is an element which intervenes between V1 and V2,pas: La Maria no ha pas aprovat l'examen 'Maria has not passed the exam'. There are also a few emphatic monosyllabic adverbs which may also occasionally intervene. Some altemative explanations are available: intervening elements are infixes; intervening elements are heads of other functional projections through which Vl+V2 must move, and a morphological reorganization of elements occurs; incorporation is not obligatory in these cases. Leaving this issue open, we must nevertheless account for the facts in (3)-(5). which represent cases of strong ungrammaticality for all Catalan speakers. The future tense in Catalan is a clear instance of incorporation leading to a totally fused head: fer + he =faré 'make + have = will make'. Notice that in this case incorporation of the infinitival forn is to the left of the auxiliary, unlike the sequences we are considering. mechanism is adjunction, as in any head-to-head movement, here it is adjunction to the right departing from other head-to-head processes. In previous frameworks there were no restrictions on the direction of adjunction, but this is a problem for Kayne (1993). We consider this in sections 3 and 5. 2. Types of Incorporation and Types of Affixes In the preceding analysis, the first verbal element of the sequence behaves like an inflectional affix, triggering the movement of V2. However, unlike all other inflectional affixes in the language it occurs before the host it is linked to, V2, and it never totally amalgamates with it. These properties may be captured and explained by referring to Roberts (1991), as in LlinL (1991). This work analyses the halva sequences as a case of substitution incorporation as opposed to adjunction incorporation. Adjunction incorporation structures (as in clitic climbing structures) have the properties of not being obligatory, of allowing excorporation (the movement out of the incorporated constituent), and of resulting in an amalgamated constituent. In substitution incorporation structures (as in V-to-I) there is a structural slot created for the incorporated element as a result of the subcategorization frame of the affix, and this involves substitution instead of a proper adjunction. The movement is required by the subcategorization frame, and there is total amalgamation. The status of the incorporation trigger as a head disallows movement out of the complex unit after incorporation, by minimality. Note that the Catalan sequence has the properties of substitution incorporation -0bligatory nature, no excorporation-, except that the affix and the incorporated verb never wholly amalgamate. In Llinas (1991) we predict this possibility by proposing that there are different types of affixes, which follow from the combination of two features: [ksyntactic] and [kmorphological]. A positive specification for the feature syntactic implies that the affix triggers incorporation, and a positive specification of the feature morphological implies that the affix and its host amalgamate. The possible combinations are: (7) syntactic morphological (a) + + (b) + - (c) - + (4 - - (7a) is the case of V-to-I amalgamated structures, (7b) is the case of the Catalan halva sequences, (7d) are neither morphological nor syntactically relevant affixes (maybe LF affixes) and (7c) are non-syntactically relevant affixes (derivational affixes). The Catalan halva affixes trigger incorporation as a result of their morphological subcategorization, as shown in (8). We will thus claim that what makes incorporation of V2 to V1 obligatory is the type of information contained in V 1 and the application of the Stray A& Filter (Baker (1988)): aprovat d 3. Kayne (1993) and Predictions on Incorporation Kayne (1993) presents a theory in which linear ordering is totally determined by phrase structure by positing that asymmetric c-command of non-terminal nodes matches linear ordering of terminals in that it is also locally linear and, thus, it has the properties of totality, transitivity and anti~ymrnetry.~ The linear ordering of terminals is established by a Linear Correspondence Axiom (LCA), which states that d(A) is a linear ordering of T, where A is the maximal set of ordered pairs (Xj, Yj) such that for each j Xj asymmetrically c-commands Yj for a given phrase marker P with T the set of terminals. A phrase marker is admissible if it has a d(A) where transitivity holds, antisymmetry is respected and it is total so that its linear ordering is specified. The linear ordering of terminal symbols in (9) is established as follows: Set A in this phrase marker is (B,D), (B,E), (B,F), (D,F) and its d(A) is (b,d), (b,f), (d,f). Its linear ordering is established as (b,d,f) as the three properties (transitivity, totality and antisymmetry) hold. With these notions, Kayne derives the constructions allowed by X' theory, including adjunction. Adjunction in his proposal always necessarily results in the precedence of the adjoined element. Take (10) as an example, corresponding to Kayne's structure (16): Transitivity : xLy & yLz -> xLz; totality : it must cover all members of ihe set (for all dlstinct x, y, either xLy or yLx ); antisymmeby: not (xLy & yLx). Consider only the adjunction part of this phrase marker. In order to establish the linear order of the terminal symbols we must first note that Kayne's definition of c-command voids segments of the ability to c-command: (1 1) X c-commands Y iff X and Y are categories and X excludes Y and every category that dominates X also dominates Y. By virtue of the fact that the adjoined head Q asymmetrically c-commands the category M -M is a segment and does not exclude Q-, the linear ordering of the two elements is established: the terminal node dominated by Q will precede the terminal node dominated by M. Note that if we invert the order of Q and M, problems arise for the LCA: In structure (12), Q asymmetrically c-commands M -again, M is a segment and it does not exclude Q-, hence, the order predicted is (q, m). Therefore, even if elements right adjoin, "... the present theory has the necessary consequence that an adjoining head will invariably precede the head it adjoins to" (Kayne (1993:27)). In other words, we cannot use right adjunction to derive a specific surface order. If we consider structure (6) again, repeated here, we are left with an impossible structure for a desired order, the adjoining element (V2) asymmetrically c-commands V1 -as Q does in (12)-and the predicted surface order is *aprovat ha , *aprovar va. Considering the more detailed incorporation structure (8), also repeated here, the prediction is somewhat different: aprovat LI In a structure such as this, the Linear Correspondence Axiom does not specify a linear ordering for V1 and V2. Notice that here, V2 cannot be claimed to asymmetrically c-command V1 -as