The Passive and Related Constructions in Marathi"

It is becoming increasingly clear among studies oflanguages ofthe world that the passive expression is related to other constructions such as the reflexive, reciprocal, spontaneous•, potential, and honorific. The passive construction in Marathi is also a case in point. Presently available characterizations of the Marathi passive cannot offer a unified account, let alone its correlations with other constructions, owing to their discrete nature and narrow perspective. In light of this, this paper offers a detailed alternative analysis of the Marathi passive construction (1) accounting for a hitherto ignored twin passive viz. the GO and COME passive, and (2) demon-strating that the passive construction in Marathi exhibits a continuum with other constructions along the lines suggested by Shibatani (1985).


l. INTRODUCTION
The passive construction has occupied a prominent position in linguistic descriptions in the last few decades, and has been analyzed by modern linguists in various frameworks, such as transformational and generative grammar (Chomsky 1965, Kuroda 1979, relational grammar (Perlmutter & Postal 1977, Pandharipande 1981, Rosen & Wali 1989, functional grammar (Givon 1979). As pointed out by Shibatani (l985), however, these formal as well as functional approaches are too restricted to account for the patterns ofdistribution which passive morphology exhibits. It has also been pointed out that various languages employ the same morphosyntactic properties as the passive in refiexive, reciprocal, spontaneous, potential and honorific constructions (Langacker & Munro 1975, Shibatani l985). The Marathi morpheme V-PERF+GO is also a case in point, and is used in four construction viz. the passive, spontaneous, potential and honorific.
Previous characterizations ofthe passive in Marathi (Tarkhadkar 1836, Kher 1899, Joshi 1900, Damle 191 1, Berntsen & Nimbkar 1975, Pandharipande 1981, Rosen & Wali 1989 are not only unable to provide a unified account, but also too narrow in perspective to account for the correlation of the passive with the other constructions. In light of this, the goal of this paper is to (a) provide an alternative account for the passive in Marathi, pointing out the inadequacies in the previous descriptions, and (b) explore the correlations of the Marathi passive with other constructions such as the spontaneous, potential and honorific. 124 Prashant PARDESHI As pointed out by Shibatani (1985), for a correct understanding of the correlation of the passive with the other constructions, a broader perspective is required. Accordingly, we will adopt the prototype framework proposed by Shibatani (1985) to explore the corre}ation ofthe Marathi passive with the other constructions. Let us first introduce the framework in which our analysis will be canied out.
1.1. TheFramework: Shibatani (1985) Shibatani adopts a traditional view of grammatical voice as a category that signals an unmarked vs. marked distinction ofmapping between the basic syntactic functions of subject and object, and the basic semantic roles of agent and patient. In an active clause, the agent occupies the most prominent syntactic slot ofsubject, and thus cannot be deleted. However, in a passive clause the agent is removed from the most prominent syntactic slot, and is either de-, moted to the role of an adjunct or not encoded at all. Passive voice, therefore, can be s'een as a means ofdemoting the agent or deleting it altogether. Thus, the prototypical passive is agentless. Shibatani (1985:837) defines the passive prototype as follows: (1) Characterization of the passive prototype Shibatani points out that, rather than arguing whether a given construction should be considered passive or not, a description must be offered as to what extent the construction in question is similar to or different from the prototype. This view ofgrammar thus assumes that various constructions exist along a continuum. Some ofthem are prototypical, others are similar to the prototype to a limited extent, while still others share no similarities at all with the prototype.
Such an approach is essential in understanding the relationships among various constructions within a single language, and is capable ofproviding a usefu1 framework for cross-linguistic research. The rationale for adopting this framework comes from the fact that other frameworks cannot offer a unified account of the passive construction or its correlation with other construc-tions. The correlations discussed here are neither purely syntactic nor semantic in nature, but are rather pragmatic--i.e. all ofthem share a common pragmatic function viz. agent defocusing .
In what follows we will demonstrate that the broader pragmatic notion ofagent defocusing offers a unified account for the passive construction in Marathi, as well as the constructions The Passive and Related Constructions in Marathi' 125 related to it. We will start our discussion with the passive, which has been extensively discussed in earlier studies.

THE MARATHI PASSIVE
Marathi has a twin periphrastic passive construction which typically consists of an agent NP, if at all present, followed by a postposition kaD Un or dwAre, and a transitive verb with either a perfect participle marker 1-1-1 followed by the verb ljA-1 `go' (hereafter referred to as the GO passive), or a participle 1-NyAti followed by the verb lye-1 `come' (hereafter referred to as the COME passive). These are exemplified below.i In her extensive survey ofpassive constructions, Siewierska (1984) points out that different languages employ different verbs as passive auxiliaries, and that many languages possess more than one periphrastic passive--e.g. a BE and BECOME passive (Swedish, Latvian, Polish, Finnish, Nez Perce), a BE and GET passive (English) or a BECOME and GO passive (Bengali)-which are not freely interchangeable. The passive construction in Marathi is a case in point, which employs GO and COME as passive auxiliaries. Among the Indo-Aryan languages, Maithili also uses GO and COME as passive auxiliaries. Siewierska further points out that the characteristics associated with passive clauses which contain particular auxiliary verbs appear to be language specific, and the use ofa given auxiliary in languages which possess more than one such constituent is determined by a variety ofsemantic, syntactic and stylistic factors.
In earlier analyses ofthe Marathi passive construction, the COME passive was either not discussed at all, or was treated as a construction synonymous with the GO passive. We will demonstrate with ample illustration that this, however, is far from true. The GO passive and the COME passive are neither interchangeable nor mutually exclusive in all contexts. Moreover, they are not discrete entities but rather form a continuum.
We shall begin with a review ofearlier analyses ofthe Marathi passive in order to establish some background, and then argue that these are in fact inadequate to account for the Marathi passive--to say nothing ofits correlation with other constructions. and modem linguistic descriptions [Berntsen & Nimbkar (1975), Pandharipande (1 98 1 ), Rosen & Wali (1989)]. For the sake ofconvenience we will revjew them separately.
As for the morphological aspect, the traditional grammarians discuss in detail the diachronic change from the suffixal to the periphrastic passive. With regards to the syntactic aspect, they define the karmaNIprayog (passive construction) as one in which the verb agrees with the object (karma [lit. deedlfatelobject]) in number, person and gender. According to this definition, whenever the subject is marked and the object is unmarked, the resultant construction would be the karmaNI, or passive construction. This definition is incorrect in that in Marathi, the verb always agrees with an unmarked nominal (ifany), and is thus independent ofany construction. Example (5) is in fact an ergative constniction, and should be distinguished from the passive constructjon.2 Example (6) is a datjve subject construction which is void of an agent.
The traditional analyses completely ignore the semantic and pragmatic aspects ofthe passive constmction, yet--except for Joshi (1900)--all have discussed the GO as well as the COME passive. Their analyses, however, have treated these constructions as semantically identical, and thus mutually interchangeable in all contexts. We claim that this is not the case, and will discuss both of these constructions in greater detail later on. The traditional grammarians also fail to capture the correlation ofthe passive with other constructions such as the spontaneous, potential and honorific.
Having summarized the traditional analyses, let us now turn to modern linguistic descriptions pertaining to the Marathi passive. Although, Berntsen & Nimbkar (1975) talk about the GO as well as the COME passive; however, they remain silent about their distribution and thus tacitly treat them as synonymous.
They also fail to recognize the correlation ofthe passive with other constructions. Pandharipande (1981) as well as Rosen & Wali (1989), on the other hand, completely ignore the COME passive. The relational grammar approach adopted by these authors is discrete in nature, and fails to capture the correlation between the passive and related constructions like the spontaneous, potential and honorific. We will demonstrate in the next section that this failure to recognize the correlations has led them to make unwarranted claims regarding the Marathi passlve.
Before proceeding further, a discussion on the differences between the passive construction and the ergative construction is in order. It should be recalled that the traditional grammarians failed to make a distinction between the ergative construction and the passive construction in Marathi [CÅí (5)]. It is thus appropriate to clarify that these constructions are not the same and therefore should be treated as independent.

The Passive and the Ergative construction
In Marathi, the subject ofa transitive clause (third person) in the past perfective is marked with the ergative marker ne (singular) or nl (plural). If the subject is a first or second person, there is no overt ergative marker; in either case, however, the verb agrees with the patient nominal, and not with the agent nominal. Note the following examples: (7) mlltUlrAm-ne noTis wAc-l-I IfYoulRam-ERG notice.F read-PAST-F `IIYoulRam read the notice.' The fact that the patient nominal ofthe ergative construction behaves like a subject (i.e. controls concord), makes it look like the passive construction. However, these two constructions exhibit fundamental differences. First, the ergative construction can be passivized, while the passive constructlon cannot: (8)rAm-kaDUn noTis wAc-1-I Ram-by notice.F read-PERF-F Example (8) is the passivized form ofthe ergative construction (7), while (9) is the `passivized' form of the passive construction (8). Note that (8) is grammatical while (9) is not.
Second, the agent nominal can be omitted in the passive, but not in the ergative construction. Deletion ofthe agent nominal in the ergative construction (viz. (7)) tums out to be ungrammatical, as shown in (1O) below. However, the deletion ofthe agent in a passive expression (viz, (8)) does not affect its acceptability, as shown in (1 1 To sum up, jn this section we have presented a critjcal review ofearlier research regarding the Marathi passive, and pointed out the inadequacies therein. We have also demonstrated that the passive and the ergative are independent constructions in their own right, and that the traditional treatment ofergative constructions as passive is inappropriate. In the next section we will present an alternative analysis of the Marathi passive construction.

AN ALTERNATIVE ANALYSIS OF THE MARATHI PASSIVE
As pointed out earlier, the previous treatments--traditional as well as modem--are unable to account for the Marathi passive and its correlation with other constructions. In this section we will present an alternative analysis ofthe Marathi passive. The analysis presented here addresses two issues viz. the distribution of the GO and COME passive, and the correlation of the'passive construction with other morphosyntactically related constructions which have not been taken up in earlier studies: the spontaneous, potential and honorific.
To begin with, we agree with Pandharipande (1981) Rosen & Wali (1989) classify the Marathi passive into two categories on the basis of the meaning they convey: regular passive (RP) and capabilitative passive (CP). Thus, like Pandharipande, they also claim that the passive expresses a capabilititative meaning.
It is true that in Marathi, the same morphology viz. V-PERF+GO is employed for expressing (non)capabilitative meaning; however, we claim that this is not the passive constructionper but rather the potential construction which shares morphosyntactic similarities with the passive.  `Even after a heavy meal one can easily eat ice-cream.' The above data thus undermines the claim that a capabilitative meaning is peculiar to the Marathi passive. As pointed out earlier, it is not the passive construction per se which expresses the capabilitative meaning, but rather a potential construction which exhibits morphosyntactic simi-larities to the passive.
According to Pandharipande, another function uniquely performed by the passive in Marathi is that ofexpressing a social convention, and thereby prescribing a particular mode ofbehavior The potential, as well as the honorific construction shares morphosyntactic similarities xN,ith the passive. To conclude, the discrete approach adopted by Pandharipande and Rosen & Wall faiis to recognize these correlations, and has led them to make unwarranted claims regarding Marathi passlves.
Let us now turn to another important issue related to the pag. sive that has not bee'n addressed in previous analyses, viz. the distribution ofthe GO and COME passives.

1 The Distribution ofGO and COME passives
As mentioned earlier, most ofthe previous studies on the Marathi passive have not even mentioned the COME passive, while the few that have treat it as semantically identical with the GO passive and thus interchangeable in all contexts. In this section we will demonstrate that the GO and COME passives are not semantically identical, and that the notion of"intention" plays a key role in their distribution.
We claim that, in Marathi, COME passives are typically employed to depict a meticulously planned, highly intentional event in which the agent plays the role ofthe planner, and who brings about the event with the aim of achieving the desired outcome. The CJO passive, on the other hand, is typically employed to depict a non-meticulously planned, less intentional event. It should be noted that the notion of"intention" is not a matter of"all or nothing" but rather a matter of"degree". The higher the degree ofintention, the greater the chances ofemploying the COME passive by the speaker and vice versa.
From the above explanation, one can predict that in the case of a meticulously planned (highly intentional) event, the GO passive will be blocked and--conversely--in the case ofa nonmeticulously planned (less intentional) event, the COME passive will be prevented from appearing. Both of these predictions are borne out, as exemplified below: The events depicted in the above examples are less intentional, and typically lack meticulous planning in that there is no definite entity that excercises conscious effort in bringing them about. Thus, as per our proposal, they can be couched only in the form ofa GO passive, while the COME passive is barred. Native speakers ofMarathi do make a distinction between GO and COME passives. Note the following contrast: (27) tyA-IA sewAjeshthate-nusAr badhatl di-1-I he-DAT seniority-asper promotion.F give-PERF-F ge-1-i /* de-NyAt A-l-I go-PAST-Fl give-PTCPL come-PAST-F `He was promoted on a seniority basis.' (28) tyA-IA sewAjeshthatA DawlUn badhatl de-NyAt he-DAT seniority violating promotion.F give-PTCPL A--1-I 1" di-1-i ge-1-i come-PAST-F lgive-PERF-F go-PAST-F `He was given promotion, violating the norms ofseniority.' 134 Prashant PARDESHI Example (27) depicts a event in which a controller who has the authority to award a promotion does so as a part ofsome routine procedure, while example (28) depicts a meticulously planned activity, overriding an otherwise routine procedure. The former depicts a typically less intentional activity, while the latter depicts a typical highly intentional one. In consonance with our proposal then, the COME passive is blocked in (27) while the GO passive is blocked in (28),

Additional evidence
In this section we will provide some additional evidence to support our proposal that the notion of"intention" dictates the selection ofGO or COME as a passive auxiliary in Marathi.

2. 1. Adverbial Modptcation
Ifour proposal is correct, adverb cukUn `mistakenly' should occur only with GO passives which typically depict non-meticulously planned, less intentional events and should be blocked in COME passives which typically depict meticulously planned, highly intentional events. Conversely, the adverb muddAm `purposely' should occur only with COME passives and should be barred from GO passives. All of these predictions are borne out as exemplified below: (29) rAm-kaDun Ram-bỳ depicts an event in which Ram did not intend to kill Ravan, butjust by mistake happened to kill him. The killing ofRavan in this case is less intentional and involves no meticulous planning and hence the GO passive is used. The situation in (30) isjust the opposite. In the case of(3 1), the COME passive is employed--which depicts a meticulously planned, highly intentional event.
Nevertheless, the adverb cukun `by mistake' which modifies it is not compatible with such an event, and consequently presents a contradiction. The situation in (32) isjust the opposite. The adverb muddAm `on purpose' creates a contradiction in the context ofa GO passive, as they do not depict meticulously plarmed, highly intentional events. As predicted by our proposal, (3 1) as well as (32) turn out to be ungrammatical. Marathi has a modal expression that conveys the notion of possibility. COME passives depict meticulously planned, highly intentional events and thus do not leave any room for speculation. On the contrary, GO passives can accommodate a speculative meaning. In consonance with our proposal, GO passives can thus accommodate the modal6s speculativelpossibilitative meaning, while the COME passives can not:

BE
It is interesting to note that ifthe speaker ofthe utterance (34) knows in advance about some plan to cheat, the utterance is perfectly grammatical.3 In such a case, the speaker uses the COME passive but at the same time does not want to commit himself, and thus wams the interlocutor in a non-committal way by using the possibilitative rnodal form. This shows that extra-linguisticl pragmatic information plays a crucial role in the grammaticality judgments ofthe GO and COME passlves.

Non-human agents
Passive expressions with non-human agents are always couched in the form ofa GO passive, and never in the form ofa COME passive. This fact receives a natural explanation under our proposal, in that non-human agents lack intention and the ability to plan meticulous p}anning in order to bring about an event. Note the following contrast: Consider further the above-mentioned examples. The nominals in the agent slots (viz. hi"uda.ỳ heart' andphuipAkhare `butterflies') are non-volitional, yet potent entities. We regard these expressions to be ambiguous between the passive and spontaneous constructions, On the agentvolitionality parameter, they can be interpreted as spontaneous, while on the agent-potency parameter they can be interpreted as passive.
To summarize, the evidence presented in the foregoing discussion lends strong support te our proposal that the notion of"intention" plays a crucial role in determining the distribution of GO and COME passives. We will now turn to the issue ofthe correlations that the Marathi passive has with other constructions (the spontaneous, potential and honorific), which had not been addressed in the previous research.

THE PASSIVE AND RELATED CONSTRUCTIONS
Having discussed the passive constniction in the previous section, we now focus on the constructions related to it, namely the spontaneous, the potential and the honorific constructions. Shibatani (1985: 825) claims that various constructions can be related not just in morphosyntactic or semantic terms, but also in tenns ofcommon pragmatic function. In Marathi, the passive, spontaneous, potential and honorific constn)ctions share morphosyntactic simi'larities in that the agent is consistently marked with an oblique marker, while the main verb is rendered in the V-PERF+GO form. We will demonstrate that despite the morphosyntactic similarities these constructions are independent in their own right, and yet they are related to the passive construction through a pragmatic notion, namely agent defocusing. Let us begin with the spontaneous constructlon.

The Spontaneous Construction
A prototypical spontaneous construction depicts an event that occurs on its own without the intervention ofan external agent. In other words, spontaneous events are void ofagency and volition. Spontaneous expressions in Marathi share morphosyntactic similarities with the passive construction--even though they differ semantically in that the former typically Iacks an agent, while in the latter an agent is always involved. This correlation finds natural explanation in the framework offered by Shibatani (1985) in that the spontaneous and the passive g. hare a common pragmatic function viz. agent defocusing. As Shibatani (1985: 838) states: Defocusing ofan agent is highly germane to spontaneous events andstates, An eventpredicated ofan agent is basicalljv causative; i, e. , an event is brought about by an agent. But an event dissociatedfrom an agent is one occurring spontaneously. Thus a sentence with a defocused agent may be utilized to describe a spontaneous event.
The Passive and Related Constructions in Marathi' 137 In spontaneous expressions the agent is absent altogether, while in the passive the agent is posited and defocused syntactically. Let us take a closer look at the spontaneous construction in Marathi.

The Spontaneous Construction in Marathi
In Marathi, passive morphology viz. V-PERF+GO is employed to express a spontaneous event as exemplified below: (37) yA kAdambarl-cl pahill don prakarNa mljANiwpurwak this novel-of first two chapters.F I consciously lihi-l-I. parantu nantar-cl prakarNe matra ml lihill write-PERF-F but after-of chapters.F however l wrote Ahet ase ml muLIc mhaNaNAr nAhl. tl lihi-1-i ' The periphrastic CAN potential construction is felicitous in both positive and negative contexts, while the AW construction is felicitous only in negative ones, GO potential constructions are generally felicitous in negative contexts, but they are not rejected altogether in positive ones. The CAN potential construction is an active construction in that the agent is unmarked. In contrast, the AW and GO potentials are non-active constructions, as evidenced by the fact that the agent exhibits oblique marking. As pointed out earlier, the discrete approach adopted by Pandharipande (l981) and Rosen & Wali (1989) (51)] would be either barred or treated as something unrelated to the passive. We disagree with both of these analyses, as they are discrete in nature and fail to provide a unified account for the phenomenon under consideration. In the potential construction, the obliquely marked agent is obligatory, whereas in the passive construction it can be optionally deleted. In all the above-mentioned examples, the agent phrases marked with kaDUn cannot be deleted--except in (52) and (53), which present cases ofoverlap between the passive and the potential construction. Under the gloss (a) interpretations, they are passives and permit agent deletion, while under the gloss (b) interpretations, they are potentials and do not permit the potential construction can be formed with intransitive, as well as transitive verbs (exemplified above). Thus the domain ofapplication ofthe potential and the passive construction is not the same, and as such they must be recognized as independent constructions. What the potential construction in Marathi shares with the passive is the pragmatic function of agent defocusing, and is thus morphosyntactically marked in a similar way.
Pandharipande (198l) also claims that AW potential constructions express (in)capabiiity of the agent, determined by agent-internal conditions like headaches, pain, hatred, happiness, physicallpsychological pain etc. GO potential constructions, on the other hand, express an (in)capability based on the agent's effort, and is determined by agent-external conditions like the weather. CAN potential constructions are neutral to whether their capability is determined by agent-intemal or agent-external conditions, and whether or not the agent expends any effort in accomplishing the task. As mentioned earlier, however, the claims made by Pandharipande regarding the cofiditions that determine an (in)capability reading are not empirically supported [Cf. (18), (19)].
The 60 potential construction is the only form that is relevant to our analysis because it shares morphosyntactic similarities with the GO passive. In subsequent discussion we thus confine ourselves to an examination ofthe GO potential construction.

DIACHRONIC DEVELOPMENT & SYNCHRONIC DISTRIBUTION
The correlations discussed in this paper are not isolated phenomena peculiar to Marathi, but are widely attested in various unrelated languages such as Japanese, Spanish, and rl"elugu.
Moreover, the correlations are not accidental--both language-internally as well as cross-linguistically--but are rather systematic. Shibatani (1997) rightly points out that in order to understand the pattenis ofsynchronic polysemy, we must know something about the historical development of the constructions in question. To substantiate this claim he discusses the diachronic aspects of voice constructions, and points out that historically middle voice forms are the major source of nk assive constructions.
In Classical Greek the middle voice category used to express reflexive, reciprocal, spontaneous and passive meanings, while in Spanish, reflexive constructions have given rise to other constructions like the spontaneous and the passive. The evolutionary path for the develor ment of the Spanish passive is then Reflexive>Spontaneous>Passive. The Japanese passive also deveJoped from a spontaneous construction. Shibatani (1998) proposes the Principle ofMaximization of Contrast as the driving force behind these diachronic changes. This principle motivates a language to develop voice constructions so that a meaning contrast is maximized. Although t(ie spontaneous construction seems to be the major source ofpassive constructions, there may be other sources too (CÅí Haspelmath 1990).
It Marathi the passive, the spontaneous, the potential and the honori fic constructions all exhibit synchronic polysemy in their morphosyntactic marking, viz. the main verb in the perfective, followed by the auxiliary GO. In order to offer an explanation for this synchronic polysemy it is necessary to probe the historical development ofthese constructions. A diachronic account is beyond the scope ofthe present study, however, and must be left for future research.

SUMMARY AND CONCLUSIONS
In the foregoing sections we have addressed a language-specific issue, viz. the treatment ofthe Marathi passive, pointing out that currently available accounts are inadequate. The alternative analysis presented here throws light on issues such as the distribution ofGO and COME passives, and the correlations ofthe Marathi passive with the spontaneous, potential and honorific constructions. Previous analyses failed to recognize these corelations, owing to their discrete approach and narrow perspective. Moreover, formal views ofgrammar (transformational, relational), have ignored the spontaneous, potential and honorific constructions from their scope ofanalysis, and have thus failed to account for the synchronic polysemy exhibited by the constructlons m questlon.
To conclude, the pragmatic notion of agent defocusing holds the key to unraveling the mystery of the synchronic relationship of the passive construction with other constructions, and thus provides strong support for Shibatani's claim that various constructions can be related not simply morphosyntactically or semantically, but also in terms ofcommon pragmatic functjon.  2 Cf. Section 2.2 for a detailed discussion ofthe passive and the ergative construction.
3 I would like to thank Ms. Vaishali Vaidya for pointing out this subtle contextual interpretation.