<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1523">
  <Title>apos;Single Cycle' Languages: Empirical Evidence for TAG-Adjoining</Title>
  <Section position="5" start_page="154" end_page="154" type="metho">
    <SectionTitle>
3 The traditional approach
</SectionTitle>
    <Paragraph position="0"> The standard approach in transformational syntactic theory since Chomsky (1965) and to this day (Chomsky, 2001) maintains that syntactic movement dependencies are a priori unconstrained by the size of the structure over which they are formed; in fact, in this approach there are no a priori restrictions on structure building at all. The structure building operation 'Merge' applies recursively until the material available for sentence building (lexical items, previously built chunks of structure) is exhausted. This approach has an inherent difficulty handling the Russian/Polish facts since it is not clear what would prevent a dependency to stretch as long as the size of the structure permits, in some languages but not others.3 The usual strategy in this case would be to impose additional constraints on movement in 'single cycle' languages which do not apply in languages like English. This may be satisfactory at some level of analysis, but involves a real complication in this theory. A more attractive possibility, we believe, would be to have this constraint follow from the  character of LDDs in English and other languages, mentioned in Section 1 is part of that difficulty.</Paragraph>
    <Paragraph position="1"> architecture of the theory itself. TAG provides just the right platform to make this explicit.</Paragraph>
  </Section>
  <Section position="6" start_page="154" end_page="154" type="metho">
    <SectionTitle>
4 A TAG solution
</SectionTitle>
    <Paragraph position="0"> We explore the linguistic version of TAG in Frank (2002) which bears close resemblance to the mainstream Minimalist model. In this version of TAG syntactic movement is naturally limited by the size of maximal structural domains built by Merge - elementary trees. Crucially, all movement takes place within elementary trees, before these trees are joined together into a complex structure by designated operations - Substitution and Adjoining.4 The recursive character of LDDs ('successive cyclicity') is seen in this system as a consequence of recursion in structure building at particular structural nodes, such as C' or T' (in the sense of X-bar theory). In particular, the recursive aspect of LDD is captured via the structure building operation Adjoining which interposes additional structure in between the head and the tail of a local dependency at a recursive node within a given elementary tree (see Section 1).</Paragraph>
    <Paragraph position="1"> Notably, in virtually all cases of LDDs considered in Frank's study the additional structure operated by Adjoining constituted a Tensed domain.</Paragraph>
    <Paragraph position="2"> This approach suggests a natural direction to pursue with respect to 'single cycle' languages that can be summarized in (10):</Paragraph>
  </Section>
  <Section position="7" start_page="154" end_page="155" type="metho">
    <SectionTitle>
(10) Proposal
</SectionTitle>
    <Paragraph position="0"> TAG-Adjoining is inoperative in 'single cycle' languages.</Paragraph>
    <Paragraph position="1"> If Adjoining is unavailable, there is no way to combine two elementary trees as in (2). (10) straightforwardly accounts for the fact that Russian and Polish feature neither A- nor A'-LDD, that is, the type of constructions in which recursive ('successive cyclic') movement is involved. This proposal makes no recourse to additional theoretical constructs as the traditional approaches but makes use of the existing machinery of TAG which provides a simple and accurate description of the phenomenon.</Paragraph>
    <Paragraph position="2"> In effect, (10) implies that a source of parametric variation lies in the phrase structural component, to which Adjoining naturally belongs. The 4Substitution connects the root node of one elementary tree in an empty slot in another elementary tree, similarly to a Generalized Transformation of Chomsky (1955/75) or Chomsky (1995), Ch.3.</Paragraph>
    <Paragraph position="3">  idea of phrase structure as a locus of parametric variation, and implications for child language acquisition and learnability, have been explored in detail in Lebeaux (1988/2000), a precursor to standard Minimalism. We believe it is possible to frame (10) in the general scheme of Lebeaux's parametric model.</Paragraph>
  </Section>
  <Section position="8" start_page="155" end_page="155" type="metho">
    <SectionTitle>
5 Parametric and acquisitional aspects
</SectionTitle>
    <Paragraph position="0"> Lebeaux (1988/2000) proposes that particular grammars are hierarchically ordered by their complexity: a grammar G0 that features operations O1 and O2 properly contain a grammar G1 that features only O1. Considering the operations Adjoina and Conjoin-a, Lebeaux represents the relevant parametric space as in (11), where arrows are to be read as addition of an operation to the grammar, and parenthesis as 'invisibility' for the learner.</Paragraph>
    <Paragraph position="2"> Different parametric options correspond to different sets of erased parentheses (outermost first).</Paragraph>
    <Paragraph position="3"> Furthermore, Lebeaux proposes that the parametric sequence (11) actually mirrors (in his terms, is 'congruent to') the time course of children's grammatical development. That is, in the course of language development children proceed from less to more computationally complex grammars, along the lines of (11).</Paragraph>
    <Paragraph position="4"> Frank (1998) takes up the developmental portion of Lebeaux's congruency thesis in the context of TAG-Adjoining, suggesting that the developmental sequence for English speaking children proceeds from the grammar without Adjoining to a grammar with Adjoining. Viewed in this manner, the proposal explains, among other things, why children learning English initially fail to construe even simple cases of long-distance wh-movement or subject to subject raising, while performing well on constructions with similar processing load that do not involve recursion.</Paragraph>
    <Paragraph position="5"> Representing Frank's proposal with Lebeaux type notation may look as in (12) (Merge and Move operate within an elementary tree; cf. above).</Paragraph>
    <Paragraph position="7"> In the context of Lebeaux's congruency thesis, Frank's proposal begs a question as to whether there exist a parametric sequence that corresponds to the proposed developmental sequence. Frank does not attempt an answer. But now we are able to fill in this gap. Specifically, we now say that, indeed, the parametric sequence includes a computationally more complex grammar with Adjoining which properly contains the grammar without Adjoining, as represented in (13).</Paragraph>
    <Paragraph position="8">  (13) Adjoining</Paragraph>
    <Paragraph position="10"> Here, one parametric option is G1 (no parentheses erased) corresponding to 'single cycle' languages like Russian and Polish. The option erasing the parentheses in (13) results in languages with usual recursive LDDs (English etc). This is exactly as expected under the Congruency thesis. 'Single cycle' languages thus provide strong evidence for 1) the TAG operation Adjoining; 2) Lebeaux's congruency thesis; and 3) Frank's acquisitional sequence with respect to Adjoining.</Paragraph>
  </Section>
  <Section position="9" start_page="155" end_page="156" type="metho">
    <SectionTitle>
6 Refining Adjoining
</SectionTitle>
    <Paragraph position="0"> Auxiliary trees, utilized by Adjoining, come in two varieties, both of which adhere to a principal requirement: the 'root' and 'foot' node of such tree must be categorically identical (e.g. CP), in order for Adjoining to succeed. In one variety the root node directly dominates the foot node (14-a). This case corresponds to standard transformational adjunction. In the second variety there is structural material between the root and the foot  The recursive structures we are interested in involve only the 'interpolation' variety in (14-b).</Paragraph>
    <Paragraph position="1"> But (10) refers to the prohibition of Adjoining in general. That is, in the present form it is too powerful: it rules out not only 'interpolated' cases of Adjoining, but also regular cases of base-generated adjunction, e.g. VP or DP modifiers (adverbs or adjectives).</Paragraph>
    <Paragraph position="2"> One direction that one might undertake in this regard is to relax (10) and allow Adjoining for particular nodes in Russian, while excluding it for others. This amounts, essentially, to specifying the list of recursive nodes for grammars of particular  languages. In this manner, we automatically constrain the types of possible auxiliary trees, targeted by Adjoining. Such lists are commonly used in various formal versions of TAG (cf. (Abeill'e and Rambow, 2000)). Our parametric variation could then be captured for instance as follows: (15) English: Aux = {TP, CP, VP, DP} Russian: Aux = {VP, DP} Another, more interesting alternative, is to make a principled distinction between the two cases of Adjoining. In fact, there is a well established linguistically sound method of distinguishing the types of root and foot nodes in (14)a and (14)b.</Paragraph>
    <Paragraph position="3"> The method goes back to structural distinction between segments and full categories, along the lines of Chomsky (1986) (who, in turn, builds on the work of R. May). Namely, both nodes labeled A in (14)a are in fact segments of a single category A. In contrast, the nodes labeled A in (14)b are full categories (note that the 'listing' solution above ignores this state of affairs). It seems appropriate, therefore, to split Adjoining into two different operations, e.g. Adjunction (which coincides with the traditional transformational usage) for (14)a, and Interpolation for the case (14)a. The proposal in (10) then pertains to the latter, without loss of generality. Details of this alternative are discussed in Stepanov (2006).</Paragraph>
  </Section>
  <Section position="10" start_page="156" end_page="156" type="metho">
    <SectionTitle>
7 Further issues
</SectionTitle>
    <Paragraph position="0"> The proposal explored in (10) does not imply that the recursive component is completely excluded in 'single cycle' languages. Declarative sentences with one or more embedded tensed clauses are of course available. In the linguistic version of TAG adopted here, those are built by Substitution - at the CP node (for details, see Frank (2002).</Paragraph>
    <Paragraph position="1"> Furthermore, wh-extraction facts concerning control infinitivals and subjunctives and Russian and Polish suggest that certain recursive structural domains (e.g. VPs in control infinitivals) are built by Merge within a single elementary tree, and therefore, that not all prima facie LDDs are exclusively handled with Adjoining, in contrast to Frank (2002). In particular, Adjoining is responsible only for LDDs that involve more than one Tense domain, while all others are built with Merge within a single elementary tree, and are not, strictly speaking, LDDs at all as they do not display the 'successive cyclic' character.</Paragraph>
    <Paragraph position="2"> This raises two further issues. One issue concerns a possible need to slightly modify the criteria of well-formedness of elementary trees formed by Merge as discussed by Frank (2002) to allow the above contexts. Another issue concerns making more precise the proper division of labor with respect to two types of LDDs. In a system such as Frank (2002) the distinction can be captured in terms of selectional restrictions, perhaps of semantic kind. Selection usually plays a crucial role in forming an elementary tree by Merge: in most recent transformational theories, selection directly determines a candidate for Merge. On the other hand, it is conceivable to suppose that Adjoining the operation that interpolates one elementary tree into another after both have already been built by Merge - has little to do with selection. Therefore, dependencies that are formed via selection in direct or indirect manner, cannot be relegated to Adjoining. Further aspects of this suggestion remain to be explored.</Paragraph>
  </Section>
class="xml-element"></Paper>