Alsina (1996) and Butt (1995) argue that complex predicate (CP) formation in Romance and Urdu, respectively, should take place in the syntax. Their analysis assumed argument structure merger and a relation between thematic arguments and grammatical functions via (different) versions of LFG's mapping/linking theory. Eschewing the use of a separate projection for a(rgument)-structure, Kaplan and Wedekind (1993) proposed a solution in terms of a Restriction Operator, which manipulates f-structure representations. A major drawback of their solution, however, was that it required lexical stipulation. This drawback has been overcome with Butt, King, and Maxwell's (2003) demonstration that it is possible to implement the restriction analysis of CPs for Urdu in such a way as to capture the original observations by Alsina and Butt satisfactorily. Wedekind and Oersnes (2003) further show that this version of restriction can be used to analyze analytic passive constructions in Danish. As with the Urdu CP, this allows for the uniformly syntactic, productive encoding of valency changing operations that affect the syntax of the language
These analyses have left open the issue of how to treat morphologically triggered valency changing alternations. In this paper, we argue that a finite-state compatible view of morphology in combination with the existing restriction analyses allows for a straight-forward analysis of Urdu causatives. The analysis is supported by the interaction of the causative with CPs and passive.
Consider how the restriction analysis works for Urdu CPs. CPs combine a light verb with a verb, noun, or adjective to produce a new verb. The Urdu permissive as in (1b) involves a composition of two argument structures: that of the main verb (`cough') and that of the light verb (`give'). Butt (1995) proposes an argument-structure analysis in which the permissive (`give') is a three-place predicate which takes an event (the main verb) as one of its arguments. The permissee argument is identified with the highest thematic role of the embedded event. From the perspective of a restriction analysis, the permissive ``adds'' a new subject and ``demotes'' the other verb's subject to a dative-marked indirect object, as in (1b), cf. (1a).
| 1a. | naadyaa | kHAAsii |
| Nadya.Nom | coughed | |
| `Nadya coughed.' | ||
| 1b. | yassiin=ne | naadyaa=ko | kHAAs-n-ee | diiyaa |
| Yassin=Erg | Nadya=Dat | cough-Inf-Obl | gave | |
| `Yassin let Nadya cough.' | ||||
| 2a. | [ PRED | 'cough<SUBJ>' | |
| SUBJ | [ PRED | 'Nadya' ] ] | |
| 2b. | [ PRED | 'give<SUBJ,'cough<OBJ-GO>'>' | |
| SUBJ | [ PRED | 'Yassin' ] | |
| OBJ-GO | [ PRED | 'Nadya' ] | |
| TNS-ASP | [ ASP | perf | |
| TENSE | pres]] | ||
The Urdu CP analysis outlined above and Wedekind and Oersnes's (2003) analysis of the Danish passive crucially rely on the syntactic, phrase-structure compositional aspects of these phenomena. In particular, there must be a c-structure node in which to put the restriction annotation that alters the valency of the verb, creating the final f-structure. However, in many languages, valency changing operations are clearly morphological: classic examples are applicatives and causatives in many languages (Alsina 1993). The question then arises as to whether restriction can be extended to provide a uniform analysis of valency changing operations.
To investigate this possibility, we examine the Urdu causative, which unlike Urdu CPs is created morphologically by affixation. There are two causative morphemes in Urdu vaa and aa. These interact with the valency of the verb and with its semantics (Saksena 1982). The surface differences in case marking, seen in (3), reflect a difference in affectedness (Saksena 1982, Butt 1998). In (3a) the causee is an accusative affected object of the main clause. In (3b) the causee is an instrumental oblique which is not affected (it is just the instrument). The basic f-structure for the non-causative verb in (3a) is shown in (4a). The basic f-structure for the resulting causative is shown in (4b).
| 3a. | anjum=ne | saddaf=ko | masaalaa | cakHvayaa |
| Anjum.F=Erg | Saddaf.F=Acc | spice.M=Nom | taste-Caus-Perf.M.Sg | |
| `Anjum had Saddaf taste the seasoning.' | ||||
| 3b. | anjum=ne | saddaf=se | masaalaa(=ko) | cakHvayaa |
| Anjum.F=Erg | Saddaf.F=Inst | spice.M=Nom(=Acc) | taste-Caus-Perf.M.Sg | |
| `Anjum had the seasoning tasted by Saddaf.' | ||||
| 4a. | [ PRED | 'taste<SUBJ,OBJ>' | |
| SUBJ | [ PRED | 'Saddaf' ] | |
| OBJ | [ PRED | 'seasoning' ] ] | |
| 4b. | [ PRED | 'cause<SUBJ,'taste<OBJ-TH,OBJ>'> | |
| SUBJ | [ PRED | 'Anjum' ] | |
| OBJ-TH | [ PRED | 'Saddaf' ] | |
| OBJ | [ PRED | 'seasoning' ]] | |
The structures in (4) can be related via the same type of restriction rules used to analyze CPs. However, restriction must take place within the formation of the lexically causativized verb. The integration of finite-state style morphologies into the LFG architecture provides a way to do this. The morphology used in the Urdu analysis associates a surface form with a lemma and a set of morphological tags. This is shown in (5a) for a simple inflected verb and in (5b) for its causative.
| 5a. | banaa | ==> | ban +Verb +Perf +Masc +Sg |
| 5b. | banayaa | ==> | ban +Verb +Cause +Perf +Masc +Sg |
The lemma and morphological tags are then parsed by the c-structure sublexical rules (Kaplan et al. 2004); these sublexical rules are formally identical to standard c-structure rules. As a result, the +Cause morpheme provides a phrase-structure locus for the restriction operator to apply to, similar to how it applied with the syntactic CPs. The relevant annotated c-structure rule is shown in (6).
| 6. | V ==> | Vstem | CauseMorph |
| !/PRED/SUBJ=^/PRED/SUBJ | ^=! | ||
| (!SUBJ)= { (^OBJ-TH) | (^ OBL) } |
There are two major benefits to the integration of this finite-state morphology plus restriction analysis of Urdu causatives. The first is that the interaction of the causative with CP formation and with passive can be directly accounted for. For example, the passive is formed periphrastically with an auxiliary based on the verb `go'. With respect to (3), the passive ``removes'' (demotes) the subject `Anjum' and promotes the former object `spice' to the subject of the clause. This periphrastically conditioned valency change can operate straight-forwardly on the result (4b) of the morphologically conditioned causative restriction operation in (6).
The second advantage is that the finite-state morphologies used in this analysis are compatible with other proposals for the integration of morphology into the formal LFG architecture. For example, Karttunen (2003) shows that Realizational Morphology (Stump 2001) which has been extensively argued to be suitable for LFG (LFG02 workshop on morphology) is finite-state equivalent.
We argue that morphological valency changing operations such as causative can be analyzed using the restriction operator. This allows for the seamless integration with other valency changing operations in Urdu. The key to the formal integration of this analysis is the interaction of the morphology with the syntax, in particular in the domain of the annotated phrase-structure rules.