File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/06/w06-1507_metho.xml
Size: 20,819 bytes
Last Modified: 2025-10-06 14:10:40
<?xml version="1.0" standalone="yes"?> <Paper uid="W06-1507"> <Title>Negative Concord and Restructuring in Palestinian Arabic: A Comparison of TAG and CCG Analyses</Title> <Section position="3" start_page="0" end_page="49" type="metho"> <SectionTitle> 1 Palestinian Arabic Negative Concord </SectionTitle> <Paragraph position="0"> In Palestinian Arabic (PA), negative concord occurs with the determiner wEla &quot;(not) even one,&quot; where negative concord describes the failure of an expression which expresses negation in some sentences to do so in others. Phrases formed with wEla (&quot;wEla-phrases&quot;) are interpreted either as negative quantifiers (&quot;NQ-wEla)&quot; or as polaritysensitive indefinites (&quot;NPI-wEla&quot;). wEla-phrases have an NQ-interpretation preceding the finite verb or verb complex in a clause (1-2) or in fragment answers (3-4): &quot;There wasn't one day the food didn't please me!&quot; &quot;The food pleased me every day.&quot; NQ-wEla never occurs within the scope of negation but does occur in post-verbal positions which are not &quot;thematically entailed&quot; by the verb (6-7)1: iVsi.</Paragraph> <Paragraph position="1"> thing &quot;She is conceited for absolutely NO reason!&quot; The NPI-interpretation is only available within the scope of antimorphic operators (Zwarts, 1993), like sentential negation or bidu:n &quot;without&quot; (8-9): &quot;I didn't give anything at all to even one of them.&quot; It follows from the distributions of NQ- and NPIwEla that wEla-phrases are blocked from post-verbal argument positions which are thematically entailed and which are not within the scope of an antimorphic operator.</Paragraph> <Paragraph position="2"> 1Following (Herburger, 2001), &quot;thematically entailed&quot; means that the meaning of the verb entails the existence of an entity filling the thematic role in question.</Paragraph> <Section position="1" start_page="49" end_page="49" type="sub_section"> <SectionTitle> 1.1 Negative Concord and Locality </SectionTitle> <Paragraph position="0"> PA negative concord is generally subject to strict locality constraints: a wEla-phrase must be contained within the smallest inflected clause containing its licensor. It cannot be separated from its licensor by the boundary of either an indicative (11) or a subjunctive/irrealis (12) complement: ) h.ada.</Paragraph> <Paragraph position="1"> one &quot;I don't think that she likes ANY one.&quot; This suggests that negative concord is a strictly bounded dependency like agreement marking, argument realization, or reflexive binding. However, there are exceptions to this generalization. &quot;Long-distance&quot; negative concord is possible between a matrix negation morpheme and wEla-phrases inside the complements of a small class of verbs, including bidd- &quot;want&quot; (15), Xalla &quot;to allow&quot; (16), h.a:wal &quot;to try&quot; (17, 25 below) or Qirif &quot;to know how to, to be able to&quot; (18 below): &quot;They wouldn't let me eat even one bite!&quot; The embedding can be recursive, provided that only verbs in this class are used (17).</Paragraph> <Paragraph position="2"> These verbs correspond to verbs found in many other languages which trigger a process often referred to as restructuring or clause union. I follow (Aissen and Perlmutter, 1983) in calling them trigger verbs. Restructuring involves the &quot;stretching&quot; of the domain of locality for certain kinds of bounded dependencies from the complement of a trigger verb to include the clause that it heads. At present no other phenomena have been identified in PA which independently confirm that it has restructuring. However, long-distance negative concord is identified as a restructuring phenomenon in several languages such as West Flemish (Haegeman and Zanuttini, 1996), Polish (Dziwirek, 1998), and Serbian (Progovac, 2000). As such, I assume for now that long-distance negative concord in PA is a form of restructuring as well.</Paragraph> </Section> </Section> <Section position="4" start_page="49" end_page="52" type="metho"> <SectionTitle> 2 A TAG Analysis </SectionTitle> <Paragraph position="0"> Restructuring involves a seeming paradox involving a dependency which is non-local in the hierarchical structure of a sentence but local in its semantics. TreeAdjoining Grammarsare wellsuited for analyzing restructuring because the distinction betweenaderived tree andthe derivation tree associated with it provides two notions of locality. Restructuring phenomena which have been analyzed with TAGs include clitic-climbing in Spanish and Italian (Bleam, 2000; Kulick, 2000), long-distance scrambling in German (Rambow, 1994), and long-distance agreement inTsez (Frank, 2006). It therefore is natural to explore a TAG analysis for long-distance negative concord in PA.</Paragraph> <Paragraph position="1"> To illustrate with a simple example, the negative concord dependency in (18) is licensed within an initial tree headed by Ektib &quot;write,&quot; and is then &quot;stretched&quot; by adjunction of the auxiliary tree headed by Qirift &quot;I was able to&quot; (19): The locality constraint on negative concord can then be expressed as a generalization about the derivation tree (20): a wEla-phrase and its licensor must be sisters:</Paragraph> <Paragraph position="3"> However, several properties of negative concord in PA preclude a simple analysis like this.</Paragraph> <Section position="1" start_page="50" end_page="50" type="sub_section"> <SectionTitle> 2.1 Clause-local Dependencies </SectionTitle> <Paragraph position="0"> The first property is the domain of locality of the negative concord dependency. In a simple TAG, syntactic dependencies are licensed within an elementary tree: they are tree-local. However, negative concord in PA is clause-local, because wEla-phrases are not licensed within the immediate tree to which they are attached, but instead within the immediate clausal tree containing them. For example, wEla-phrases can be inside prepositional phrases attached to a negative clause (21-22): &quot;You don't disagree with us about even one thing.&quot; In a simple TAG analysis, the wEla-phrase first substitutes into the initial tree headed by the preposition, which is then attached to the clausal tree. The relationship between the wEla-phrase and its licensor would therefore not be tree-local.</Paragraph> <Paragraph position="1"> Clause-locality can be modeled with what I refer to as &quot;Scope TAG&quot; (Kallmeyer and Joshi, 2003), a multi-component TAG (MC-TAG) in which quantificational NPsare tree sets containing two parts: a &quot;defective&quot; auxiliary tree IP* which specifies the scope of the quantifier, and an NPtree which specifies its restriction. I refer to such tree sets as &quot;scope sets.&quot; While Kallmeyer & Joshi's proposal is intended to capture the semantic scope of quantifiers, it can alsobe used toexpress clause locality byassigning PPs to scope sets as well, and by stipulating that scope sets can combine with each other by means of set-local adjunction. TheIP*-node inthe scopeset of a wEla-phrase can then adjoin to the IP*node in the PP scope set, which in turn adjoins to the IP-node of the initial tree.</Paragraph> <Paragraph position="2"> For example, (21) above can be analyzed with the elementary trees in (23) (trees are in abbreviated form), producing the derivation tree in (24):</Paragraph> <Paragraph position="4"> However, given (24) it is still not possible to state a generalization about negative concord locality in terms of sisterhood in the derivation tree.</Paragraph> <Paragraph position="5"> This can be remedied by adopting the &quot;nodesharing&quot; relation proposed by (Kallmeyer, 2005). Informally, two nodes a and b are in a node-sharing relation in a derivation tree T iff they are either in a mother-daughter relation in T at a node address A, or there is a sequence S of nodes N1 ...Nn which is the transitive closure of a mother-daughter relation in T in which the node pairs are related in terms of the root node or foot node in an auxiliary tree.</Paragraph> <Paragraph position="6"> On this basis, the negative concord locality generalization is that a wEla-phrase and its licensor are &quot;shared-node sisters&quot; in the derivation tree, where shared-node sisters are two nodes A and B which are each in a shared-node relation with a single node C. For example, in (24) b is a shared-node parent of both a1 and d. Accordingly, a1 and d are shared-node sisters with respect to b.</Paragraph> </Section> <Section position="2" start_page="50" end_page="51" type="sub_section"> <SectionTitle> 2.2 Trigger Verbs and Complement Type </SectionTitle> <Paragraph position="0"> The second property of PA long-distance negative concord that complicates a TAG analysis has to do with the kinds of complement that they take.</Paragraph> <Paragraph position="1"> TAG approaches to restructuring exploit &quot;reduced complement&quot; analyses in which trigger verbs take &quot;smaller&quot; complements than other kinds of subordinating verbs do (Bleam, 2000; Kulick, 2000).</Paragraph> <Paragraph position="2"> However, PA trigger verbs are mixed in terms of the types of complements they take: h.a:wal &quot;try to&quot; or k.idir &quot;be able to&quot; optionally allow a complementizer Pinn- (25-26), while bidd- &quot;want&quot; or Qirif &quot;know to, be able to&quot; exclude it (27-28): Assuming that the presence of a complementizer indicates a CP category, and that the presence of agreement marking onthe verb indicates anIP category, what these data show is that some trigger verbs allow either CP or IP complements, while others allow only IP complements. It follows that complement category cannot be exploited as away todistinguish trigger verbs from non-trigger verbs.</Paragraph> <Paragraph position="3"> This is an essential distinction because restructuring is not the only phenomenon which involves adjunction. For example, long-distance A-dependencies are analyzed in TAG as involving adjunction of auxiliary trees. (29-30) show that the same verbs which block long-distance negative concord allow long-distance A-dependencies, indicating that theymust alsobe analyzed as auxiliary trees. Moreover, (30) can include the complementizer Pinn-, indicating that it takes the same kinds of complements as do trigger verbs like k.idir &quot;be able&quot; and h.a:wal &quot;try&quot;: &quot;What did you promise to give her?&quot; A failure to distinguish between trigger verbs and non-trigger verbs will over-predict the availability of long-distance negative concord.</Paragraph> <Paragraph position="4"> To make this distinction, I use Dowty's (Dowty, 1994) analysis of negative concord licensing.</Paragraph> <Paragraph position="5"> Dowty models negative concord with a &quot;polarity&quot; feature whichtakes &quot;+&quot; or&quot;-&quot; values. Whenanegative concord item combines with a clausal category it specifies (by unification) the clause as having a negative value for this feature. In addition, Dowty assumes that root clauses must have a positive value for the feature: I refer to this as the root clause polarity constraint. Negation morphemes (as well as bidu:n &quot;without&quot;) take a complement specified as POL- and return a constituent with a POL+ feature. A root clause containing a negative concord item and lacking a negation morpheme will have a POL- feature for its root node and violate the root clause polarity constraint. This derives the requirement that wEla phrases in root clauses be &quot;roofed&quot; by a negation morpheme.</Paragraph> <Paragraph position="6"> Turning to long-distance negative concord, trigger verbs can be distinguished from non-trigger verbs by stipulating that non-trigger verbs take POL+ complements, while trigger verbs (and auxiliary verbs) impose no polarity specification and instead inherit the polarity feature withwhichtheir complement is specified2. An analysis of this kind applied to (18) would result in a derived tree (32) which satisfies the root clause polarity constraint.</Paragraph> </Section> <Section position="3" start_page="51" end_page="52" type="sub_section"> <SectionTitle> 2.3 Negation Morphology </SectionTitle> <Paragraph position="0"> The last property of long-distance negative concord sentences to be dealt with has to do with negation morphology inPA.Negation is expressed with some combination of the proclitic ma:- and the enclitic -s. -s appears to be a second-position attaching to the first word-sized constituent in the stringproduced byanIP-constituent, provided that the word contains a morpheme expressing person features (Awwad, 1987; Eid, 1993).</Paragraph> <Paragraph position="1"> The most frequent distribution has -s attached to the leftmost verb stem in a clause, which may be the main verb in a mono-verbal predicate (33), or to the leftmost auxiliary in a clause with compound tense-aspect-mood marking (34-35): In other kinds of sentences, -s attaches to a variety of non-verbal expressions, including the indefinite pronoun h.ada &quot;(any)one&quot; (36), the existential particle fi: (37), inflected prepositions (38), and the adverb QUmr &quot;ever&quot; (39): &quot;There are people who have never posted a thread on almontada.com.&quot; What these expressions all have in common with verb stems is that they occur as the first constituent in the clause and that they all contain a morpheme expressing person features. It follows that -s is constrained to occur in the second position attached to a word that is inflected for person.</Paragraph> <Paragraph position="2"> The cases in which -s attaches to a verb can be modeled by assuming that ma:- and -s are part of a tree set and that -s adjoins to right of an I-node: The cases with -s attached to a non-verbal expression require a second analysis. One possibility is to assume a second tree for -s like the first, except with -spreceding the foot node. This requires stipulating a morphological output filter that affixes -s tothe preceding word and blocks use ofd2 in(40): This is still not adaquate for (35), in which -s is attached to a &quot;serial auxiliary&quot; (Hussein, 1990), one of a small set of verb stems which function as aspectual adverbs and which &quot;agree&quot; with the main verb in aspectual form and agreement marking. Serial auxiliaries are plausibly analyzed as The structure resulting from (44) has two I-nodes, and another constraint would have to be stipulated forcing -s to adjoin to the leftmost of the two.</Paragraph> <Paragraph position="3"> To sum up, a TAG analysis can be formulated for PA long-distance negative concord which allows the locality of negative concord licensing to be stated as a generalization about shared-node derivation trees. However, the analysis requires brute force stipulations to capture the morphological expression of negation in PA negative sentences. Moreover, the TAG analysis does not provide a way to express the simple morphological generalization that -s falls in the second position in the string generated by the clause.</Paragraph> </Section> </Section> <Section position="5" start_page="52" end_page="54" type="metho"> <SectionTitle> 3 A CCG Analysis </SectionTitle> <Paragraph position="0"> The TAG analysis has difficulty accommodating the distribution of -s because TAG trees are phrase structures, making it difficult to state constraints on strings of words rather than on hierarchical structure. Categorial Grammar, on the other hand, is a string calculus, and its operations result in string concatenation rather than structure expansion. For this reason, a CG can be constrained to not generate particular kinds of strings, rather than particular trees. A CG therefore provides a way to state constraints on the distribution of -s more directly than a phrase-structure grammar does.</Paragraph> <Paragraph position="1"> I assume a Combinatory Categorial Grammar (Steedman, 1996; Steedman, 2000b; Baldridge, 2002). The basis of the CCG analysis is that npI-wEla-phrases are treated as type-raised categories which look for an s category to their left. I continue following Dowty in assuming the root clause polarity principle and in assuming that wEla-phrases specify a POL- feature on the s-headed category that they combine with.</Paragraph> <Paragraph position="2"> NQ-wElaphrases, on the other hand, are treated as negative quantifiers which look for their s-headed argument to the right: The -s morpheme fixes a clause with a POLfeature, while ma:- takes the POL- clausal category and changes its value for the polarity feature to POL+, satisfying the root clause polarity constraint. This works much as the TAG analysis did. The slash in the type for -s is marked with the &quot;crossed composition&quot; modality. This allows -s to combine withapreceding s-headed category while returning a category looking for its arguments to the right (Figures 1-2)4.</Paragraph> <Paragraph position="3"> Turning to long-distance negative concord, a CCG analysis, like the TAG analysis above, has to account for the distinction between trigger verbs and non-trigger verbs. The CCG analog of auxiliary-tree adjunction is function composition. The long-distance negative concord dependency therefore involves a specific kind of composition subject to stricter constraints than is the more general kind which produces A-dependencies.</Paragraph> <Paragraph position="4"> In order to model this, I adapt Hepple's (Hepple, 1990) approach tomodeling island constraints in Categorial Type Logic. In brief, Hepple's approach is to assign unary modalities to the arguments of clausal categories (such as subordinating verbs or relative pronouns) as well as to the nominal argument of a type-raised extracted category (such as a question word or topicalized noun phrase). The former are referred to as &quot;bounding modalities,&quot; and the latter as &quot;penetrative modalities.&quot; Interaction axioms require the penetrative modality of an extraction category to be compatible with the bounding category of its argument in terms of a type hierarchy defined over modalities.</Paragraph> <Paragraph position="5"> The unary modalities in CTL can be duplicated in CCG as features on category labels, so to approximate Hepple's proposal, I define a feature hierarchy as follows: (51) h g c Each pair of sisters in the hierarchy consists of a &quot;penetrative feature&quot; and the &quot;bounding feature&quot; which blocks it (following Hepple's terminology).</Paragraph> <Paragraph position="6"> The feature c is an penetrative feature which is blocked by the g feature, and h is the most general or permissive bounding feature.</Paragraph> <Paragraph position="7"> The idea is that categories which participate in restructuring dependencies are marked with the c penetrative feature, which is spread across all the arguments of a complex type: (52) wEla h.ada :- Sc$\(Sc$/NPc) Trigger verbs impose the h bounding feature on their complements, while non-trigger embedding verbs impose the g feature: (53) bidd- &quot;want,&quot; Qirif &quot;be able to,&quot; h.a:wal &quot;try to&quot; :S\NP/(Sh\NPh) null (54) waQad-yu:Qid &quot;promise to&quot; :- S\NP/(Sg\NPg) According to (51), categories marked with feature h are compatible with categories marked with feature c, while categories marked with feature g clash with it. The clash between g and c expresses the restriction on restructuring dependencies.</Paragraph> <Paragraph position="8"> For example, in an analysis of (18), wEla kilmi applies to the composed constituent, Qirift Ektib. This is possible because the penetrative feature c on the wEla-phrase is compatible with the h bounding feature which Qirift passes to its complement (Figure 3).</Paragraph> <Paragraph position="9"> Long-distance negative concord is blocked in two ways. A wide-scope derivation (in which the wEla-phrase combines with the composition of the matrix and embedded verbs) is blocked by a feature clash between the g and c features (Figure 4). A narrow scope derivation (in which the wEla-phrase combines with the embedded verb only) is blocked because of a resulting clash in polarity features between the embedded clause and the matrix verb (Figure 5).</Paragraph> </Section> class="xml-element"></Paper>