Feature-based argument mapping and animacy optimization in impersonal passives.
Analysis of the alternation between impersonal reflexives and reflexive passives in Spanish shows that higher animacy makes an argument less marked as an object. This generalization supports Hopper and Thompson's (1981) "transitivity hypothesis", since it is entailed by their claim that objects tend to be highly individuated. The article proposes an optimality-theoretic implementation of the analysis, using a feature-based approach to argument realization modeled on Bresnan and Kanerva's (1989) lexical mapping theory. The proposal highlights a contradiction between the transitivity hypothesis and the approaches to animacy that follow Silverstein's (1976) analysis of split ergative languages, for which alternatives are suggested.
1. Two constructions with arbitrary human agents in Spanish
1.1. Impersonal reflexives and reflexive passives
The use of a reflexive pronoun (or a related form) to encode the occurrence of an arbitrary subject is common among the Slavic, Romance, and Germanic languages (Siewierska 1984, 1988). (2) In Spanish, a reflexive clitic pronoun se marks the suppression of an arbitrary human subject. This arbitrary se, as I will call it, can occur with intransitive verbs, as in (1a), or with transitive verbs, as in (1b). These are examples of impersonal reflexives.
(1) impersonal reflexive
a. Los domingos no se trabaja.
The Sunday.PL not SE work.3SG
'One does not work on Sunday.'
b. Se felicito a los soldados.
SE congratulate.3sG to the soldiers
'The soldiers were congratulated.'
In Spanish, the patient in an arbitrary se construction can be realized as an object or as a subject. In this article I will show that the alternation is conditioned by the semantic features of the patient, animacy in particular. The way in which animacy factors into argument realization alternations and morphosyntactic phenomena like case marking and agreement is an issue of central concern for functional-typological approaches in linguistics. One influential proposal is the transitivity hypothesis of Hopper and Thompson (1981). They suggest that transitivity is a property of a clause, and not of individual predicates. Animacy is one of the ingredients that, through a category of individuation, contributes to transitivity: animate arguments are more individuated than inanimate ones, and the more individuated an object is, the more transitive the clause is. In this article I argue that the realization of the patient in the arbitrary se construction gives support to Hopper and Thompson's hypothesis, since inanimate patients are realized as subjects (Siewierska 1984; Hidalgo 1994), but patients higher in animacy are realized as objects. (3)
The role of animacy in the characterization of the marked and the unmarked object is a source of controversy in the functional-typological literature, since it has also been suggested that arguments high in animacy are marked as objects, in contradiction to Hopper and Thompson's claim. (4) This hypothesis is originally proposed in Silverstein's (1976) work on split ergative systems, and is clearly stated by Comrie (1981): "The most natural kind of transitive construction is one where the A is high in animacy and definiteness, and the P is lower in animacy and definiteness; and any deviation from this pattern leads to a more marked construction" (Comrie 1981: 121). Aissen (1999) formalizes Silverstein's hypothesis in optimality theory, to show that it correctly accounts for some voice phenomena that are sensitive to animacy. But her proposal predicts that inanimate patients should be realized as objects in the arbitrary se construction (and animate patients as subjects), which, I will argue, is the opposite of what the data show. This construction, then, makes it apparent that the functional accounts of animacy effects on argument realization based on Hopper and Thompson (1981) and on Silverstein (1976) are incompatible with each other. In this article I conclude that the approach based on Silverstein's claims should be abandoned. I develop an optimality-theoretical analysis of arbitrary se based on Hopper and Thompson's hypothesis instead, which also incorporates elements of the lexical mapping theory of Bresnan and Kanerva (1989). To account for some of the facts discussed in Aissen (1999), I propose that a system of relational constraints should be adopted instead of the markedness constraints she employs.
The article is structured in the following way: In the rest of Section 1 I discuss previous analyses about the structure of the arbitrary se constructions, to conclude that they display an alternation in the realization of the patient, between an object and a subject. In this section I also show that the animacy features of the patient determine its grammatical function, in accordance with Hopper and Thompson's transitivity hypothesis. In the remainder of the article I develop an optimality-theoretical implementation of this analysis. In recent years, a growing body of research (Aissen 1999; Bresnan 2000) has found in optimality theory (Prince and Smolensky 1993) a suitable framework to formalize many principles of functional linguistics. One of the goals of this article is to do the same for aspects of the transitivity hypothesis. In Section 2 I lay down some fundamental concepts of OT, in particular their application to syntax, and I present my OT implementation of an argument realization theory based on the lexical mapping theory of Bresnan and Kanerva (1989). In Section 3 I apply this theory to explain the animacy effects on the arbitrary se construction. I also propose a functional motivation for the analysis, based on Hopper and Thompson's (1981) transitivity hypothesis, and I show that the rival approach based on Silverstein's (1976) and Aissen's (1999) ideas makes the wrong predictions for the Spanish data. Some alternative proposals to capture the broad effects of animacy on argument realization are discussed in this section, lexical decomposition grammar in particular (Wunderlich 1997; Stiebels 2000). The conclusion is that while OT has the flexibility to express different functional analyses, the real theoretical discussion still centers around the nature and content of the constraints.
1.2. Structural differences between personal and impersonal passives
In Spanish, arbitrary se can occur in more than one construction. When a sentence with arbitrary se is subjectless and in the 3rd person singular, I refer to it as an impersonal reflexive. (5) If the verb in the impersonal reflexive is transitive, as in (lb), the clause has an overt direct object (notice the overt case marking and the lack of agreement with the verb). I will refer to the object of sentences like (lb) as the PIVOT. The pivot can also be realized as a subject, as in (2). In this case the pivot controls verbal agreement and has nominative case. When the pivot is a subject, the construction will be referred to as a reflexive passive. To refer to impersonal reflexives and reflexive passives as a class I will use the term arbitrary se construction.
(2) reflexive passive
Se arreglaron las sillas.
SE fixed.3PL the chairs
'The chairs were fixed.'
Cinque (1988) develops a proposal about the structure of arbitrary se constructions in the framework of the government and binding theory (Chomsky 1981). To formalize the opposition between the reflexive passive and the impersonal reflexive, he distinguishes between argument se and nonargument se. The reflexive clitic is adjoined under INFL in both constructions. Argument se absorbs an external theta role. Since the subject position of the reflexive passive is nonthematic, an underlying object can move there to receive nominative case (in line with Burzio's generalization, the predicate cannot assign Accusative case because there is no external theta role, having been absorbed by argument si). If the verb is intransitive, an empty pleonastic pronoun is inserted. The representation of a reflexive passive is then as in Figure la. Nonargument se, on the other hand, does not absorb an external theta role. The verb's theta roles are assigned in normal fashion. The subject of an impersonal reflexive is, in Cinque's analysis, an empty pronoun that bears a thematic role, more similar to a null personal pronoun than to an expletive. The syntactic structure of an impersonal reflexive is presented in Figure lb. (6)
The distinction between the reflexive passive and the impersonal reflexive is also defended in Dobrovie-Sorin (1994, 1995). She puts into question Cinque's analysis in terms of a contrast between argument se and nonargument se, claiming instead that the distinction between the two arbitrary se constructions is a matter of Case. In reflexive passives se is an accusative clitic, while in impersonal reflexives se is (or absorbs) nominative. She argues that nominative se licenses an empty pronoun in subject position. Accusative se, on the other hand, is a link in a binding chain that connects the subject to the object. The representations for the two arbitrary se are presented below:
(3) a. Nominative se: [pro.sub.i] [se.sub.i] V (NP)
b. Accusative se: [NP.sub.i] [se.sub.i] V [e.sub.i]
[FIGURE 1a OMITTED]
Following the analyses in Cinque (1988) and Dobrovie-Sorin (1994, 1995), I characterize the two arbitrary se constructions found in Spanish as involving an altemative realization of the pivot (the patient, roughly speaking) as an object or a subject, i.e., as an accusative argument or a nominative argument. What drives the alternation in the function of the pivot is a change in the properties of the reflexive clitic. This change, I suggest, is better described in functional terms. In the impersonal reflexive the clitic functions as the subject of the clause, performing the duties of a nominative argument. In the reflexive passive the clitic ceases to be associated with this grammatical function. The pivot can then be realized as a subject, as it must, bearing nominative case. One advantage of this approach over Cinque's or Dobrovie-Sorin's is that the semantics of the reflexive clitic remains the same throughout the arbitrary se constructions, denoting an arbitrary human actor. In the other analyses, on the other hand, se bears the index of an arbitrary human actor only in the impersonal reflexive, but for no apparent reason loses this property in the reflexive passive.
1.3. Animacy, transitivity, and the arbitrary se construction
The reflexive passive and the impersonal reflexive are not in free variation. The animacy features of the pivot determine its realization as a subject or an object. If the pivot is definite and animate, as in (1b), the impersonal reflexive construction is employed. But if the pivot is definite and inanimate, then the reflexive passive is preferred (2). It is important to note that the altemation is not determined by the lexical semantics of the predicate. As the examples in (4a)-(5b) show, the same animacy effects can be observed with verbs like atrapar 'catch', which can take either animate or inanimate objects.
(4) a. Ayer se atrapo a los ladrones.
Yesterday SE caught.3SG to the thieves
'The thieves were caught yesterday.'
b. Ayer se atraparon las pelotas
yesterday SE caught.3PL the balls
'Yesterday, the balls were caught.'
(5) a. * Ayer se atraparon los ladrones.
Yesterday SE caught.3PL the thieves
'The thieves were caught yesterday.'
b. * Ayer se atrapo las pelotas
yesterday SE caught.3SG the balls
'Yesterday, the balls were caught.'
A syntactic characterization of the arbitrary se constructions alone cannot account for these animacy effects. Animacy constraints on argument realization are discussed and analyzed at length in the functionalist literature (Silverstein 1976; Givon 1976, 1979; Comrie 1977, 1980a, 1980b; DeLancey 1981). Following Hopper and Thompson (1981), I argue that an animate pivot is realized as an object because animate objects have a higher degree of transitivity than inanimate ones, and this correlates with the higher degree of morphosyntactic transitivity of the impersonal reflexive versus the reflexive passive. Hopper and Thompson analyze transitivity as a property of an entire clause (not just of a particular subcategory of verbs), which characteristically involves an agent performing an activity that affects a patient. But for them the number of participants in an event is only one among many parameters of clausal transitivity. Different parameters of transitivity, each of which has a morphosyntactic correlate, can be isolated. Clauses can be characterized as being more or less transitive, depending on the number of high or low transitivity features they possess. The different parameters with their high and low features are summarized in Table 1.
The impersonal reflexive is higher in transitivity than the reflexive passive, having an accusative argument in its structure. A higher degree of morphosyntactic transitivity, Hopper and Thompson claim, always correlates with a higher degree of semantic or lexical transitivity. In the case of the arbitrary se constructions, the key semantic property is individuation of O. Higher transitivity is assured by clearly distinguishing the patient from the agent, and also from the patient's background. An action can be transferred more effectively to a more individuated patient. The semantic features in the left column of Table 2 contribute to a higher degree of individuation of the object, while the ones in the right column decrease the individuation of the object.
Crucially for my analysis, animate arguments are more individuated than inanimate ones. Since a higher degree of individuation of the object requires a higher degree of morphosyntactic transitivity, then it follows that an animate pivot will be realized as an object (with accusative case), while an inanimate pivot will not, occurring in the construction with reduced transitivity instead. According to the transitivity hypothesis, then, inanimate objects are more marked than animate objects. The reflexive passive avoids this marked configuration by allowing the inanimate pivot to be realized as the subject. This is possible, I suggest, because of the nature of the reflexive clitic that expresses the arbitrary human agent. In the impersonal reflexive the clitic bears the subject grammatical function, but in the reflexive passive it is associated with a different one, much in the same way as the subject of a transitive sentence can also be realized as an oblique (a prepositional phrase headed by por 'by' in Spanish) in periphrastic passives. I will delay discussion of the formal status of the reflexive clitic in reflexive passives until Section 3.1. For now, Let's assume that subject is the unmarked grammatical function for the reflexive clitic. Under normal circumstances, then, the impersonal reflexive is an unmarked construction. Why should the grammar of Spanish allow the more marked construction (i.e., the reflexive passive) to occur, then? The answer is that an inanimate object is more marked than a reflexive clitic that does not bear the subject grammatical function. If the object is animate, on the other hand, it is not marked, so there is no need to repair the clause by mapping the pivot onto the subject and causing the clitic to be marked instead.
The alternation between the impersonal reflexive and the reflexive passive, then, can be explained as the result of a conflict between two constraints, one favoring the assignment of the subject function to the reflexive clitic, another one penalizing objects with certain semantic features. When the latter is given priority over the former (that is, when the pivot is inanimate), the result is the reflexive passive. In recent years, optimality theory, or OT (Prince and Smolensky 1993), has emerged as a new linguistic framework to deal with analyses like this one, in which structures can violate certain constraints and still be grammatical because they are doing so to satisfy other constraints with a higher priority. Unlike standard generative grammar, which is operational in nature (a procedure applies to an input to produce an output), OT is comparative: the output is chosen among a set of candidates. Candidates are compared (and selected) with respect to a set of ordered, violable constraints. OT developed at first as a theory of phonology, but recent work has extended the principles of OT to syntax (Legendre et al. 1993; Grimshaw and Samek-Lodovici 1995; Grimshaw 1997; Bresnan 2000). OT syntax replaces the rules and hard principles of mainstream generative grammar with a set of violable constraints on sentence structure. Grammatical sentences may violate some of these constraints, but they are selected as the optimal output for a given input if they fare better than rival candidates with respect to higher-ranked constraints. (7) In the next sections I take advantage of this framework to present a more explicit analysis of the alternation between the reflexive passive and the impersonal reflexive.
2. Optimality theory and the analysis of argument realization
2.1. Argument realization in an OT framework
Several frameworks have been developed within OT to address the issues of argument realization. Arguments are mapped onto (or linked to) grammatical functions according to systematic tendencies and principles, but there are also predictable mismatches. OT offers an account in terms of rankings of violable constraints linking a semantic role to a grammatical function. One hypothesis about the systematic gaps in argument structure realization assumes the existence of a hierarchy of semantic roles, and also of a hierarchy of grammatical functions. Levin and Rappaport-Hovav refer to this hypothesis as the prominence preservation hypothesis. They characterize it as the claim that "the relations of semantic prominence among arguments as encoded in a thematic hierarchy should be maintained in the syntax" (Levin and Rappaport-Hovav 2005: 142). According to the PPH, agents are mapped onto the subject and patients onto the object in transitive clauses because the patient is outranked by the agent in the semantic hierarchy, and therefore cannot bear a more prominent syntactic function than the agent.
Aissen (1999) captures the effects of the PPH in an OT framework using the technique of harmonic alignment to formulate a universal subhierarchy of linking constraints. Given a binary scale of grammatical relations (6a) and a scale of thematic roles (6b), the harmonic alignment of the two scales is as in (7a). This scale gives rise to the constraint subhierarchy in (7b)
(6) a. Su > Oj
b. ag > pat
(7) a. Su/ag > Su/pat
b. *Su/pat >> *Su/ag
The constraint ranking in (7b) is universally fixed. In no language will the alignment of a subject with an agent be more marked than the alignment of a subject with a patient. This guarantees that agents, and not patients, are linked to the subject in the output of a transitive sentence, as in Figure 2. But there are cases, however, in which a higher-ranked constraint forces a violation of *Su/pat. In passives, for instance, a constraint that forces topics to be aligned with the subject (*Su/x, after Aissen 1999) favors a candidate that incurs otherwise fatal violations of the most prominent constraints in (8b). (8) This is shown in Figure 3 (X indicates high thematic prominence, x low thematic prominence).
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
2.2. The "lexical mapping theory"
In the OT approach to argument realization sketched above, then, the unmarked alignments between semantic arguments and grammatical functions emerge after a set of candidates is evaluated by a hierarchy of violable constraints. This approach captures the effects of the PPH, since these constraints are the result of a harmonic alignment of a scale of grammatical functions and a scale of semantic roles. Constructions like passive, in which a marked alignment is observed, are selected as the output when a higher-ranked constraint trumps some alignment constraint that is critical for the selection of an otherwise unmarked candidate. This model takes full advantage of the essential features of the OT framework: globality, parallelism, and violable constraints.
An alternative approach to the PPH is based on a system of ordered linking rules that map semantic roles onto grammatical functions. Levin and Rappaport-Hovav (1995) develop an "operational" version of this approach. Arguments are mapped onto syntactic positions inside or outside the VP, depending on their lexical-semantic features. (9) In case of a potential conflict between a rule that maps an argument to an internal position and another rule that maps the same argument to an external position, an extrinsic order between the rules resolves the issue. Extrinsic rule ordering is also used in some feature-decomposition approaches to argument realization, like Bresnan and Kanerva's (1989) lexical mapping theory. In this theory, arguments are classified with a system of features to determine their grammatical function. Features are monotonically assigned to underspecified arguments by rules of different kinds. In case of conflict, an earlier rule takes precedence over a subsequent one. In this section I will implement some of the results of these operational theories in an optimality-theoretical framework that replaces extrinsic rule ordering with a ranked system of mapping constraints.
My proposal will be based on Bresnan and Kanerva's (1989) lexical mapping theory (see also Alsina and Mchombo 1989; Bresnan and Zaenen 1990). They analyze grammatical functions in terms of two binary features: [r] for semantically restricted arguments, and [o] for objective. Subjects and obliques are [-o], subjects and objects are [-r]. The resulting classification is shown below.
SUBJECT OBJECT OBLIQUE RESTRICTED OBJ.
[FIGURE 4 OMITTED]
These features are used to constrain the mapping of semantic roles onto grammatical functions. Semantic roles start fully underspecified for the functions they can be mapped onto, until intrinsic features are linked to them. Once intrinsic features are assigned, default role classifications fill in some of the underspecified values. Lastly, two well-formedness principles, presented in (8a) and (8b), resolve any uncertainties or deficiencies in the assignment of grammatical functions.
(8) a. Subject condition: Every lexical form must have a subject.
b. Function-argument biuniqueness: In every lexical form, every expressed lexical role must have a unique syntactic function, and every syntactic function must have a unique lexical role.
In a transitive clause, for example, the assignment of intrinsic and default roles to agent and patient yields an agent specified to link to the subject (agents are inherently [-o], and I-r] by default). (10) The patient, on the other hand, is intrinsically [-r], but it is underspecified for the value of [o]. Function-argument biuniqueness solves the uncertainty, mapping the patient onto an object. This is shown in (9a). In some cases, however, patients will be mapped onto subjects. This is what happens in passive clauses. To account for this, a morpholexical rule assigns the feature [+r] to the agent. Since agents are intrinsically [-o], they must be mapped onto an oblique. The patient, then, is realized as the subject to satisfy the subject condition, as shown in (9b).
(9) a. transitive: (ag, pat) inherent: [-o] [-r] default: [-r] S O/S WF: S O b. transitive: (ag, pat) inherent: [-o] [-r] passive: [+r] default: Obl O/S WF: Obl S
2.3. An OT version of the "lexical mapping theory"
Feature-decomposition approaches to argument realization like LFG's linking theory manage to capture complex generalizations about argument realization, and also the reasons behind systematic mismatches between semantic role and grammatical function (or morphological case), in a monostratal model. The theory is to a great extent motivated by the observations (valid crosslinguistically) that the function of the agent alternates between subject (unmarked) and oblique (as in the by-phrase of an English passive), and that patients alternate between subject (as in some intransitives and in passives) and object (when something else is the subject). The decomposition of grammatical functions into features makes it possible to group them into natural classes. Mapping constraints can then make reference to the classes defined by these features, instead of listing arbitrary sets of grammatical functions. For instance, assigning the [-r] intrinsic feature to patients limits the class of functions to which they can be assigned to the [-r] ones: subjects and objects. After that, it is up to the well-formedness conditions, then, to decide which of the [-r] functions a patient has for a particular verb.
Implementing the lexical mapping theory in the OT framework is a very natural thing to do, since it has some optimality-theoretical mechanisms already built into it. The well-formedness conditions in the lexical-mapping theory, for instance, can be understood as constraints that evaluate candidates in which an argument (the patient in the examples discussed above) can be realized as an object or a subject, discarding the candidate that violates one of the conditions, and choosing the other one as the output. Borrowing an idea from the OT treatment of case in Woolford (2001), I interpret the well-formedness condition that requires the presence of a subject as the effect of high-ranked constraints penalizing structures with grammatical functions other than subject. These markedness constraints are presented in (10a) and (10b).
(10) a. * [+o]: The value of the [o] feature in the output is not '+'.
b. * [+r]: The value of the [r] feature in the output is not '+'.
These constraints make sure that a candidate with a subject will be more optimal than any candidates without a subject. The constraint * [+o] penalizes candidates with objects. * [+r], which rules against candidates with obliques, is not really crucial to the argument I am developing here, so it will be omitted from the tableaux. The other well-formedness condition, argument-function biuniqueness, enters the constraint set as UNIQ. (11)
(11) UNIQ: Outputs do not have more than one syntactic function for each semantic role, nor do they have more than one semantic role for each semantic function.
Besides the well-formedness conditions, however, other aspects of the lexical mapping theory are at odds with the architecture of OT. In particular, the use of default rules in the assignment of features to underspecified arguments has to be modified. In OT, the constraint hierarchy does much of the work that defaults and underspecification do in other frameworks. In underspecification analyses, structures need not be lexically specified for those features that are predictable from the context. Special rules fill in the missing values, and where they do not apply a default rule comes in to make sure the final structure is fully specified. In the lexical mapping theory, for instance, agents are underspecified for the feature [r]. A special rule assigns it the value [+] in passive sentences, and the value [-] by default (since the agent is inevitably the highest role). In OT, default features are no more than features favored by low-ranked constraints. (12) These constraints select an output with elsewhere features when competing candidates are equally harmonic with respect to some higher ranked constraints. A corollary of this approach is that candidates are always fully specified. In the OT implementation of the lexical mapping algorithm I propose, then, default rules and morpholexical rules are replaced by markedness constraints, appropriately ranked.
Lastly, an OT implementation of a lexical mapping theory must also have a way to account for the assignment of intrinsic features. I will assume that intrinsic features are already specified in the input. Preserving those features in the output is the task of a set of faithfulness constraints (see Kiparsky 2001 for similar ideas involving case features). Arguments are fully specified for the features [o] and [r]. Some examples are given below.
[FIGURE 5 OMITTED]
These features assignments define an unmarked grammatical function for each semantic role. Given a certain set of arguments in the input, there will be one candidate that is the most faithful one with respect to the features [o] and [r]. However, the constraints that replace the morpholexical operations, the default feature assignment rules, and the well-formedness principles of the standard lexical mapping theory may select unfaithful candidates. If the faithfulness constraints penalizing those candidates are outranked by some markedness constraints, then the output will be unfaithful to the input. Faithfulness to the intrinsic roles that are assigned to agents and patients in the lexical mapping theory must therefore be guaranteed by very highly ranked constraints in the OT model I propose. These constraints, which require that a feature of the input be expressed in the output, are presented in (12a) and (12b).
(12) a. MAX [-o]: A feature [-o] in the input is also in the output (addresses the assignment of the intrinsic feature [-o] to the agent).
b. MAX [-r]: A feature [-r] in the input is also in the output (addresses the assignment of the intrinsic feature [-r] to the patient).
A few tableaux will serve to illustrate the operation of a feature decomposition theory of argument realization in the OT framework. Figure 6 shows that, when the input is an intransitive verb with an agent, the output maps this argument onto a subject. Since agents have the same feature specifications as the subject function, a candidate with a subject is the most faithful one, as well as the least marked.
[FIGURE 6 OMITTED]
If the single argument is a patient, on the other hand, it bears a marked feature [+o]. The unfaithful candidate with a subject is selected as the output, however, because the faithful candidate violates the markedness constraint * [+o]. The faithfulness constraint MAX[-o] has no effect on the competition. This is shown in Figure 7.
[FIGURE 7 OMITTED]
In an active transitive clause, the most faithful candidate is the optimal one. Even though the active violates the markedness constraint * [+o], the passive incurs a fatal violation of Max[-r], since the agent is assigned a [+r] feature in the passive. This is sown in Figure 8. Passives are selected when the active does not abide by the constraints relating information structure to grammatical function, *Su/x (cf. footnote 8 for Aissen's 1999 use of this constraint). If the patient is more thematic than the agent, then, a high ranking *Su/x favors the passive, in spite of the violation of MAX[-r]. This is shown in Figure 9. To show more clearly how the constraints that I propose eliminate other candidates, I have included one candidate in which the functions assigned to agent and patient are reversed, and another one in which both roles are mapped onto the subject function. The faithfulness constraint MAX[-o] rules against the candidate with role reversal, while UNIQ disqualifies the candidate with multiple subjects.
[FIGURE 8 OMITTED]
[FIGURE 9 OMITTED]
3. An OT account of arbitrary se
3.1. The OT analysis
The optimality-theoretical implementation of the feature-based mapping theory outlined in the previous section can account for the alternation between the reflexive passive and the impersonal reflexive I developed in Section 1.3. In this analysis, the alternations are the product of a tension between two constraints, one against having clitics that are not subjects, and another one (sensitive to the animacy features of the patient) against clauses with direct objects. In a nutshell, the impersonal reflexive violates * [+o], but the rival candidate (the reflexive passive) violates a higher-ranked constraint requiring the clitic to be mapped onto the subject. This constraint is outranked by another one penalizing [+o] arguments if they also have the feature [inanimate], therefore favoring the reflexive passive. This section presents this analysis in detail, with its functional motivations in the transitivity hypothesis of Hopper and Thompson (1981), and then compares it to Aissen's (1999) approach to animacy.
The effect of * [+o]is that a patient will be mapped onto a [-o] function if no other argument is realized as the subject. An impersonal reflexive violates this markedness constraint, but it can nevertheless be the winning candidate when the rival candidate (the reflexive passive) violates a higher ranked constraint. In the reflexive passive, the agent (the arbitrary human argument, expressed in the reflexive clitic) is not mapped onto the subject function. But what sort of grammatical function should it bear? In a periphrastic passive, where the agent is not realized as a subject either, the value of its [r] feature is set to [+]. I propose to account for the realization of the arbitrary human agent in the reflexive passive in a similar way, assigning the clitic a restricted function. The reflexive passive, then, is unfaithful to the feature [-r] of the agent. Since MAX[-r] dominates * [+o], the impersonal reftexive wins in the competition shown in Figure 10.
[FIGURE 10 OMITTED]
For the reflexive passive to be the output, on the other hand, a constraint penalizing candidates with objects must dominate MAX[-r]. Since the reflexive passive is favored by inanimate pivots, I propose to make * [+o] sensitive to the semantic features of the argument that bears it. An inanimate [+o] argument is more marked than an animate one. This constraint, which is more stringent (and therefore dominates) the general markedness constraint * [+o], will be abbreviated as * [+o]/inan. It dominates the faithfulness constraint MAX[-r], selecting the reflexive passive as the winner when the patient is inanimate (if the patient is animate, on the other hand, the constraint * [+o]/inan has no effect, and the competition reverts to the one in Figure 10).
[FIGURE 11 OMITTED]
The highly marked nature of a patient bearing the [+o] feature when inanimate, then, determines that the optimal candidate is the one that is unfaithful to the [-r] feature of the agent. This is why the reflexive passive occurs when the pivot is inanimate. The question that arises at this point is why doesn't an inanimate patient make a periphrastic passive more optimal than the corresponding active. Both clause types are grammatical in Spanish, as the examples in (13a) and (13b) show.
(13) a. Lopez arreglo las sillas.
'Lopez fixed the chairs.'
b. Las sillas fueron arregladas por Lopez.
'The chairs were fixed by Lopez.'
The solution to this problem, I suggest, can be found in the different case properties of nouns and the reflexive pronominal clitic. Nouns and noun phrases must bear case--this is the Case Filter of Government and Binding theory (Chomsky 1981). I will introduce a constraint, CASE, that is violated by candidates in which one or more NPs are caseless (see Grimshaw 1997 and Wunderlich and Lakamper 2001 for a similar constraint). Nouns receive case in two ways. Structural case is assigned by the verb to its direct object (Accusative) or by Tense to the subject (Nominative). Other arguments receive inherent case, often from a preposition. In a Spanish periphrastic passive, the agent bears a restricted grammatical function, and cannot receive structural case. A preposition must be inserted to satisfy its case requirement. But this comes at a cost, since a structure with a preposition is more complex than a structure without it. In OT, the constraint that favors candidates with less structure is *STRUC. A passive sentence in which the agent is realized as the complement of a preposition incurs one more violation of this constraint than alternative structures that do not have this preposition. But without it, a [+r] agent cannot get case. If *STRUC and CASE dominate * [+o]/inan, then, an inanimate object alone cannot force a periphrastic passive to be the winning candidate, and an active structure is the optimal output when the agent is not a clitic. This is shown in Figure 12.
[FIGURE 12 OMITTED]
Clitics, on the other hand, are subject to a weaker case requirement. This is true of the reflexive clitic in particular, as revealed by research on their defective nature with respect to their function, content, and morphological properties. (13) Like nouns, clitics may bear structural or inherent case. In French for instance, the clitics en and y are used to refer to partitive and locative arguments, respectively. A historical remnant of the locative clitic can be found in the Spanish verbal form hay 'there exists (<has)'. But unlike nouns, clitics pronouns do not require a preposition to receive Case. In the case of an arbitrary human subject, then, the [+r] clitic can satisfy the CASE constraint without incurring a violation of *STRUC. With such an input, then, the effects of * [+o]/inan become apparent.
3.2. Feature-based mapping in OT and the "transitivity hypothesis"
The analysis presented above captures the insight that the alternation between the reflexive passive and the impersonal reflexive is the result of a tug between two requirements. One disfavors clauses with direct objects, the other expresses a preference for mapping agents onto subjects. Only in cases where the former trumps the latter does a reflexive passive emerge as the optimal candidate. This happens when the patient is inanimate. An inanimate object is more marked than an animate object, to the point that avoiding an inanimate object is more important than assigning the agent a function other than subject. In the OT analysis I have presented in the previous section, this is expressed in the ranking * [+o]/inan >> MAX[-r] >> *[+o ].
The subhierarchy * [+o]/inan >> * [+o], I argue, is universally fixed. It is functionally motivated by the transitivity hypothesis of Hopper and Thompson (1981), which I introduced in Section 1.3. Since [+o] defines the natural class of grammatical functions that complement the verb (direct objects and thematically restricted objects), it is a contributing feature of morphosyntactic transitivity. An OT implementation of the transitivity hypothesis, then, will associate this feature with any of the semantic features [f.sub.[-ind]] that contribute to a reduced individuation of the object (listed on the right hand column of Table 2) to produce a set of markedness constraints like * [+o]/abstract, * [+o]/plural, etc. All of the markedness constraints of the form * [+o]/[f.sub.[-ind]] are universally ranked above * [+o], since an object that has any [f.sub.[-ind]] feature is more marked than other objects. It may be that in some grammars, some constraint of the form * [+o]/[f.sub.[-ind]] dominates a constraint that favors a candidate with an object, but that this constraint is ranked above * [+o]. Every constraint that * [+o]dominates, on the other hand, is also dominated by * [+o]/ [f.sub.[-ind]]. A consequence of this is that, across languages, candidates whose objects have semantic features that do not contribute to argument individuation will tend to lose against candidates in which that argument is not realized as an object. This tendency is part of Hopper and Thompson's transitivity hypothesis, and it is accounted for in the OT implementation I propose.
An interesting consequence of the transitivity hypothesis is that some of the features that contribute to the individuation of the object can combine to increase the transitivity of a clause. Animacy may join definiteness to create an argument that is even more individuated than an indefinite one. The effects of this can be observed in the arbitrary se construction. In the examples I have discussed so far, the pivot is definite, in addition to being animate. If an animate pivot is indefinite, however, its decreased individuation makes it a candidate to be realized as a subject, as in (14).
(14) Se felicitaron soldados.
SE congratulated.3PL soldiers
'Soldiers were congratulated.'
In the OT analysis I propose, this is the result of having a constraint * [+o]/indef dominating MAX[-r], the faithfulness constraint that militates against the reflexive passive. This makes an indefinite object marked enough to force the clitic out of the subject function. The combined effect of * [+o]/inan and * [+o]/indef, then, is that MAX[-r] is only strong enough to keep definite animate arguments as objects in the arbitrary se construction.
Animacy and definiteness are independent of each other, but other features that contribute to individuation are not. Many studies about the effects of semantic features on argument realization (Silverstein 1976; Givon 1976, 1979), propose a natural scale of animacy, with inanimate common nouns at the bottom and first or second person pronouns (also referred to as the local persons) at the top, immediately followed by third person pronouns and proper names (24).
(15) Local > 3rd > Proper > Human > Animate > Inanimate.
There is a correlation between the scale in (15) and individuation: the more animate an argument is, the more individuated it is. According to the transitivity hypothesis, then, an argument with a higher degree of animacy is more likely to be associated with the morphosyntactic features of transitivity. This hierarchical effect can be observed in the arbitrary se construction, since proper nouns and pronominal arguments are excluded from the reflexive passive (16c), but they can appear in the impersonal reflexive (16a)-(16b). (14)
(16) a. Se felicito a Lopez y a Gomez.
SE congratulate.3SG to Lopez and to Gomez
'Lopez and Gomez were congratulated.'
b. Se los felicito a ellos.
SE 3PL.ACC congratulate.3sg to them
'they were congratulated.'
c. *Se felicitaron Lopez y Gomez / ellos
SE congratulate.3pL Lopez and Gomez/they
'Lopez and Gomez/they were congratulated.'
When pronouns and names that refer to inanimate entities are considered, however, the results may be unexpected. As Comrie (1981: 188) points out, "a pronoun whose referent is low in animacy is actually placed higher than a noun phrase whose referent is high in animacy. (...) The pronoun/nonpronoun opposition in fact cross-cuts the human/nonhuman or animate/inanimate opposition." Common nouns referring to geographical landmarks occur in the reflexive passive, behaving like other inanimate nouns (17a), and so are the corresponding proper names (17b), even though their occurrence in the impersonal reflexive construction is only marginally bad. (15) Inanimate-referring pronouns, on the other hand, occur in the impersonal reflexive (17c), but are excluded from the reflexive passive (17d).
(17) a. Se bombardearon/*bombardeo las ciudades.
SE bombed.3PL/ bombed.3SG the cities
'The cities were bombed.'
b. Se bombardearon/?bombardeo Granada y Guernica.
SE bombed.3PL/ bombed.3SG Granada and Guernica
'Granada and Guernica were bombed.'
c. Se las bombardeo
SE 3PL.ACC bombed.3SG
'They were bombed.'
d. * Se bombardearon ellas
SE bombed.3PL they
'They were bombed.'
The constraint * [+o]/inan takes care of the distribution of inanimate proper nouns, but pronouns referring to inanimate entities require a different treatment. Pronominal arguments are always highly individuated, regardless of whether they refer to animate or inanimate entities. This corroborates the claims made in the functional literature. One implementation of this idea, found in Silverstein's (1976) approach to animacy, is to limit the animate/inanimate opposition to nominal arguments, not pronominal ones. The constraint * [+o]/inan is not violated by pronominal objects, then, since they are never specified for animacy features. (16) As Croft (1990: 116) points out, however, this amounts to replacing a hierarchy of values (of a single feature: animacy) with a hierarchy of features. Space limitations prevent me from going any further into the theoretical consequences of this approach.
3.3. An alternative account of animacy effects in OT
The OT analysis of the alternation between the reflexive passive and the impersonal reflexive I have proposed is based on a markedness constraint that penalizes the association of a morphosyntactic feature [+o] with the semantic feature [inanimate]. The ranking of this markedness constraint relative to MAX[-r] and * [+o]determines the realization of the pivot. This analysis implements, in the OT framework, Hopper and Thompson's transitivity hypothesis, according to which low individuation reduces the markedness of an object. But, as mentioned in the introduction, there is a certain amount of disagreement in the functionalist literature about what constitutes a marked value for an object in terms of animacy. Silverstein (1976) argues that high animacy is the marked value for objects, not the unmarked one (as suggested by Hopper and Thompson). Silverstein's hypothesis is originally motivated by an implicational generalization about split ergative languages, to which I will return in Section 3.4. This hypothesis is also understood as the motivating force behind some alternations in argument realization. Comrie (1981), for instance, after discussing the effect of agreements constraints on voice, says that there is "a more general phenomenon, found in some languages, whereby voice must be used to bring a noun phrase higher in animacy into subject position" (Comrie 1981: 185). An OT implementation of this approach is developed in Aissen (1999). Using the OT principle of harmonic alignment discussed before, she derives a universal subhierarchy of animacy constraints, which allows her to explain animacy effects on voice marking and voice alternations in several languages. In this section I will show that Aissen's constraint hierarchies make the wrong predictions regarding the effect of animacy on the alternation between the impersonal reflexive and the reflexive passive. The claim that high animacy increases the markedness of an object, then, cannot be supported in the face of the negative evidence the arbitrary se construction provides. The constraints based on Hopper and Thompson's transitivity hypothesis, I will conclude, provide a better explanation.
One of the goals of Aissen (1999) is to account for the effect of animacy scales on passives (and inverse marking) in several languages. Her data cover the topmost ranks of the animacy scale in (15), that is, the person scale that places local persons (speaker and addressee) over third person, bur her model has implications for the rest of the animacy hierarchy as well. In Lummi (Straits Salish), for instance, the person of the arguments of a transitive verb determine the voice of the clause (Jelinek and Demers 1983). An active sentence cannot have a third person agent and a local (first or second) person patient. The passive must be used instead. Aissen argues that in a case like this one the active is ruled out by a constraint against local person objects. Even though the passive is a marked construction, it is less marked than the alternative. If the object is third person, on the other hand, both the active and the passive are possible, since avoiding a third person object is not a priority (the information status of the arguments determines voice instead).
The generalization that emerges from this chart, Aissen argues, is that local person objects are more marked than third person objects. Aissen's OT implementation of this analysis consists of the constraint subhierarchies in (18a)-(18b), which are the result of harmonic alignment of a binary scale of grammatical functions and Silverstein's animacy scale. In these subhierarchies, the idea that subjects are more marked the more inanimate they are, and that objects are more marked the more animate they are, receives an explicit formulation.
(18) a. Csu: * Su/inan >> * Su/anim >> * Su/hum >> * Su/3 >> * Su/loc
b. Coj: * Oj/loc >> * Oj/3 >> * Oj/hum >> * Oj/anim >> * Oj/inan
These constraints play a role in the selection of active versus passive in a language like Lummi. Passives are ruled out by a markedness constraint *Su/pat ('sentences do not have patients realized as subjects'). Insertion of this constraint at the appropriate point in the subhierarchies in (18a)(18b) yields the constraint ranking in (19).
(19) Lummi: * Oj/loc >> * Su/pat >> * Oj/3
In a competition between an active and a passive sentence, the active normally wins unless there is another constraint that dominates * Su/pat and is violated by the passive. In the case of Lutarei, Aissen suggests, that role is played by the constraint that penalizes the association of a local person with an object (* Oj/loc).
Aissen's perspective on the association between objects and animacy features, then, is that the more animate an object is, the more marked it is. This claim is diametrically opposed to my analysis, according to which an animate object is less marked than an inanimate object. The conflict between the two postures can be seen clearly in the predictions they make for the realization of the pivot in the arbitrary se constructions in Spanish. An analysis that employs Aissen's constraints predicts that the inanimate pivot should remain as an object, and that the reflexive passive should be preferred with the animate pivot. This prediction is clearly wrong. To show how this analysis works, I will assume that reflexive passives are ruled against by a markedness constraint * Obl/[se.sub.arb] (the arbitrary human clitic is not an oblique). For the inanimate pivot to be realized as a subject, the constraint * Oj/inan must dominate the constraint * Obl/[se.sub.arb]. This is shown in Figure 13. As in Aissen's analysis of animacy effects in Lummi, a candidate that violates a constraint against a passive (in this case * Obl/[se.sub.arb], the constraint that rules against the reflexive passive) can win if a rival candidate (i.e., an impersonal reflexive) incurs a disqualifying violation of a constraint against associating a semantic feature with the object (i.e., * Oj/inan). In Aissen's approach, however, the ranking * Oj/anim >> * Oj/inan is universally fixed. This means that if * Oj/inan dominates * Obl/[se.sub.arb], SO will * Oj/anim. If the input is an animate patient, this ranking will wrongly select the reflexive passive as the output, as in Figure 14.
[FIGURE 13 OMITTED]
[FIGURE 14 OMITTED]
Aissen's animacy constraint subhierarchy implements the generalization that if a semantic feature f can be associated with a direct object, any semantic feature lower than f in the animacy scale can be associated with an object as well. This is, in essence, Silverstein's hypothesis. The distribution of the pivot in the arbitrary se construction shows that this implicational generalization is invalid, which is why an analysis based on Aissen's constraints fails to account for the facts. The analysis I defended in this article, on the other hand, implements the opposite implicational generalization, based on Hopper and Thompson's transitivity hypothesis: if an object can be associated with a semantic feature f, then any semantic feature higher than f in the animacy hierarchy will be associated with the object as well. This generalization is supported by the arbitrary se facts in Spanish. My analysis, then, does not require any further stipulations to account for the alternation between the impersonal reflexive and the reflexive passive in Spanish, and thus should be preferred to one based on the claim that objects are more marked the more animate they are.
3.4. Alternative analyses of voice alternations and split ergative systems
The failure of the Silverstein-Comrie-Aissen approach to account for the animacy effects on the arbitrary se construction in Spanish has to be considered against the body of empirical evidence that seems to support the view of objects lower in animacy as less marked. The evidence comes from voice alternations that are sensitive to animacy (like the Salish data discussed before), and also from split ergative languages. To shore up my claim that higher animacy makes an object less marked, I will offer alternative analyses for voice alternations and ergative splits that do not rely on Silverstein's hypothesis.
I will start by reviewing animacy effects on voice alternations like the ones discussed in Comrie (1981) and Aissen (1999). What they find is that in some cases there is a tendency for patients high in animacy to be realized as subjects of passives instead of objects of active clauses. Aissen's analysis of Lummi, which I summarized in the previous section, is sufficient illustration of this approach: passives in Lummi are chosen to avoid clauses with local person objects, which, according to Silverstein's hypothesis, are the most marked ones. There is, however, another way to look at animacy effects on voice in Lummi: passives are chosen so that a clause does not have a subject that is lower than the object in animacy. That is, the animacy constraint on argument realization in Salish is relational. Active clauses with 3rd person subjects and local person objects violate the relational animacy constraint, while the passive counterparts do not, and that is why passives are optimal candidates. Clear evidence for the relational nature of the animacy constraints on voice systems comes from an additional restriction on active sentences. Jelinek and Demers (1983) state that Salish sentences with common noun subjects and third person pronominal objects are ungrammatical. A passive is the only option in this case. This alternation cannot be explained on the basis of a high-ranking constraint against third person pronominal objects, since these are allowed when the subject is also a pronoun. A relational constraint penalizing objects higher than subjects in animacy, on the other hand, accounts for the facts without losing generality. When the marked values of the object with respect to its animacy features are considered, then, the constraint hierarchy * [+o]/inan >> * [+o]is the one that is universally valid. Aissen's constraint hierarchy in (19b) is not empirically supported or functionally motivated, and it should be replaced by adequate relational constraints. (17)
A relational constraint does not entail any absolute markedness value for the object in terms of animacy, so it is not in contradiction with the hypothesis that an object that is low in animacy is marked. It is true that a model that includes both constraints will experience some internal tension regarding the more natural expression of the object, but this is precisely what decades of research on the morphosyntax of direct objects has uncovered. Hopper and Thompson (1981) discuss a similar problem in relation to definiteness, a semantic feature that is also relevant for argument realization. Definite objects are more individuated that indefinite ones. According to the transitivity hypothesis, then, the more definite an object is, the more likely it is to be expressed in a transitive construction. This hypothesis, Hopper and Thompson notice, conflicts with a claim made by Comrie (1977), who suggests that indefinites are more likely to be realized as objects than definites. Hopper and Thompson revise Comrie's analysis, suggesting that the more adequate generalization is a relational one: an object cannot be more definite than a subject. If relational constraints are included in the grammars of natural languages, many puzzling facts about voice, agreement, and case marking can be explained more naturally.
I will now turn to ergative/accusative splits. Silverstein (1976) notices that many languages show a split between ergative/absolutive case marking patterns and nominative/accusative patterns based on animacy. The splits are captured by a pair of implicational generalizations: If an object is marked with accusative case in a particular language, then any NP higher in the animacy scale will also be marked accusative. If a subject is marked ergative in a particular language, then any NP lower in the animacy scale will be marked ergative as well. Since accusative and ergative are morphologically marked (i.e., usually expressed by nonzero suffixes), an economy principle dictates that they should be used to flag the unexpected (hence marked) clusters of animacy and function. Silverstein concludes that objects lower in animacy are less marked than objects higher in animacy, and that these marked values are reversed for subjects.
Aissen (1999) models the economy principle behind split ergative languages in OT. Each constraint in the animacy/function hierarchy is locally conjoined with a constraint penalizing zero morphological exponence (*0). The constraint penalizing morphologically complex expressions, *STRUC, is inserted at the cut-off point for morphological marking in the hierarchy. So, for instance, a language in which pronominal objects are marked accusative, and nominal objects absolutive, will include a subhierarchy like (20). This ensures that when a human object is not marked for case, those objects that are below it in the animacy hierarchy will have to be unmarked for case as well.
(20) ... *0 &* Oj/Hum >> *STRUC >> *0 & * Oj/3 ...
Aissen's approach has its critics. Sells (2001) points out that nothing in the model rules out local conjunction of *STRUC with other markedness constraint subhierarchy. This may result, for instance, in a language that case marks objects lower in animacy while withdrawing case marking from those higher in animacy. An alternative account of ergative splits, also within an OT framework, is developed in Stiebels (2000). She suggests that accusative arguments are less marked when they are higher in animacy, and that ergative arguments are less marked when they are lower in animacy. In Stiebels' analysis, following common practice in lexical decomposition grammar (Wunderlich 1997; Wunderlich and Lakamper 2001), a pair of binary features ([[+ or -]hr] for 'there is/isn't a higher role' and [[+ or -]lr] for 'there is/isn't a lower role') cross-classic semantic roles.
Cases are partially specified for the same features. A ranked set of correspondence constraints select an optimal assignment of cases to a predicate's arguments for a particular array of thematic roles. To ensure that accusative marked arguments are at the high end of the animacy hierarchy (and ergative marked ones at the low end), Stiebels proposes the constraint subhierarchies in (21a) and (21b), which results from harmonic alignment of the animacy scale with the binary scale [+hr] > [+lr].
(21) a. * [+lr]/Loc >> * [+lr]/3 >> ... >> * [+lr]/Inan
b. * [+hr]/Inan >> * [+hr]/Anim >> ... >> * [+hr]/Loc
A candidate with an accusative (i.e., a [+hr] feature) argument incurs a more severe violation of the markedness constraint against accusative case if the argument is a human common noun than a pronoun. Clauses with accusative human arguments, then, are more marked than those with accusative pronominal arguments. Therefore, if a language with a split ergative system based on animacy marks a particular argument as accusative, it will mark any argument higher in animacy with accusative as well. (18)
Stiebels' insight about animacy and split ergative systems suggests a possible account within Hopper and Thompson's transitivity hypothesis. Assuming that accusative-marked objects are an indication of increased transitivity, then the higher an object is in the animacy hierarchy the more likely it is to be marked accusative. Any object below the cutoff point in the animacy scale for a particular language is unmarked. If a language places a priority on assigning different cases to subjects and objects, however, it can satisfy this constraint by marking subjects with ergative case. In this way, the generalization that ergative/absolutive marking is associated with arguments lower in the animacy hierarchy (and nominative/accusative with arguments higher up) can be accounted for without assuming that objects are more marked the more animate they are. By pairing this analysis of ergative splits with a relational animacy constraint on voice alternations, then, I am able to address major empirical objections against an account of animacy effects based on the transitivity hypothesis.
In this article I have presented an OT analysis of a particular alternation in argument realization in the grammar of Spanish. When the arbitrary se construction has a transitive verb, the patient can be realized as an object or as a subject, depending on its semantic features: inanimate arguments are realized as subjects, animate arguments as objects. The alternation is the result of a conflict between two constraints, one penalizing structures with inanimate objects (* [+o]/inan), another one penalizing structures in which the arbitrary se clitic is not the subject (MAX[-r]). The right candidate is selected as the output by the ranking * [+o]/inan >> MAX[-r]. This analysis is functionally grounded in Hopper and Thompson's transitivity hypothesis. According to this hypothesis, objects are marked when they have semantic features that do not increase their degree of individuation (inanimate, nonspecific, etc.). This claim is implemented in OT by means of a universal ranking that places the constraint * [+o]/inan above * [+o], a markedness constraint that penalizes candidates with objects in general. The features [o] and [r], and the way they operate in mapping arguments to grammatical functions, are taken directly from Bresnan and Kanerva's (1989) Lexical mapping Theory.
The OT implementation of Hopper and Thompson's transitivity hypothesis I proposed in this article makes different predictions than a proposal by Aissen (1999) with respect to the way animacy features determine argument realization. Since Aissen's proposal is an implementation of the functional theories of Silverstein (1976) and others, the contradiction between my analysis of animacy in the arbitrary se constructions and an analysis based on Aissen's constraints reveals a clear conflict at the heart of a functional approach to the syntax-semantics interface. This is a serious problem for a theory of the effect of animacy features on morphosyntax. A possible way out of this conflict is to say that both hypotheses are right. Universal grammar makes two markedness hierarchies available in principle to all languages. A particular language may choose the higher values of the animacy hierarchy as the marked ones for the object, or the lower values instead. This kind of typological switch is, after all, not unheard of. Languages diverge from one another in many respects that are diametrical opposites (head first or head last, head-marking or dependent-marking, etc.). This answer, however, is unsatisfactory. Unlike formal typological constraints, the animacy hierarchy is functionally motivated. The very relevance of an animacy scale for linguistics has to be grounded in the cognitive or communicational salience of the semantic features. If either endpoint can be accepted as the marked value for the object function, then that means that the scale is arbitrary, and that therefore there is no motivation for it. This begs the question of why should languages care about animacy in the first place.
My solution to the conflict is to reject as a valid linguistic hypothesis a universal subhierarchy of constraints that makes it more marked for an object to have a feature higher up the animacy scale. The phenomena that this subhierarchy is designed to account for can (and in fact should) be explained by relational constraints instead, stating that an object cannot be more animate than a subject. Admitting relational constraints into linguistic theory raises a series of formal issues that I cannot address here, and should be left for further research. But more crucially, my suggestion addresses the concern, raised in a recent scholarly exchange (Newmeyer 2000; Bresnan and Aissen 2002), of how much of our understanding of syntax is advanced by looking for a functional motivation for linguistic constraints. As Newmeyer (2002) puts it:
Nobody understands or, in the foreseeable future, is likely to understand the full set of external factors that might combine to account for the properties of syntactic structure (...). The plausible external motivations are so numerous, so diverse, and so open-ended that any conceivable rule or constraint in any framework could be provided with one. (Newmeyer 2002: 56)
Newmeyer's criticism is directed at recent work in OT syntax (e.g., Aissen 1999) that seeks a functional motivation for the constraint hierarchy. By proposing a functionally motivated class of constraints that are in direct contradiction with Aissen's, I seem to be giving Newmeyer more reasons to doubt the enterprise of Functional OT. But this is not the case. My proposal avoids the conflict with Aissen's approach by replacing her hierarchies with relational constraints. Even if a functional justification can be found for Aissen's analysis, this does not justify the inclusion of those constraints in linguistic theory. Moreover, the fact that Aissen's hierarchy may not stand alone as theoretically acceptable brings up the question of what external motivations are likely to have an impact on grammar. That is, even if it is plausible to think of an external reason that would lead a grammar to classify animate objects as less marked than inanimates, there are no constraints that do so independently of the relative animacy features of the object. From this point of view, then, the goal of functional OT should not be to provide constraints with functional motivations, but rather to understand the consequences of a particular analysis for our understanding of how language works within the context of human cognition and human interaction.
University of California, Davis
Received 29 November 2005
Revised version received
6 January 2006
Aissen, Judith. 1999. Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17. 673-711.
Alsina, Alex & Sam A. Mchombo. 1993. Object asymmetries and the Chichewa applicative construction. In Sam A. Mchombo (ed.), Theoretical Aspects of Bantu Grammar, 17-45. Stanford, CA: CSLI.
Bonet, Eulalia. 1995. Feature structure of Romance clitics. Natural Language and Linguistic Theory 13. 607-647.
Bresnan, Joan. 2000. Explaining morphosyntactic competition. In Mar Baltin & Chris Collins (eds.), The Handbook of Contemporary Syntactic Theory, 11-44. Oxford: Blackwell.
Bresnan, Joan & Jonni M. Kanerva. 1989. Locative inversion in Chichewa: A case study of factorization in grammar. Linguistic Inquiry 20. 1-50.
Bresnan, Joan & Annie Zaenen. 1990. Deep unaccusativity in LFG. In Katarzyna Dziwirek, Patrick Farrell & Errapel Mejias-Bikandi (eds.), Grammatical relations: A cross-theoretical perspective, 45-57. Stanford, CA: CSLI.
Bresnan, Joan & Judith Aissen. 2002. Optimality and functionality: Objections and refutations. Natural Language and Linguistic Theory 20. 81-95.
Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris. Cinque, Guglielmo. 1988. On si constructions and the theory of arb. Linguistic Inquiry 19. 521-581.
Comrie, Bernard. 1977. Subjects and direct objects in Uralic languages: A functional explanation of case marking systems. Etudes Finno-Ougriennes 12. 5-17.
Comrie, Bernard. 1980a. Inverse verb forms in Siberia: Evidence from Chukchee, Koryak, and Kamchadal. Folia Linguistica I. 61-74.
Comrie, Bernard. 1980b. Agreement, animacy, and voice. In Gunther Brettschneider & Christianne Lehman (eds.), Wege zur universalien Forschung, 229-34. Tubingen: Gunther Narr.
Comrie, Bernard. 1981. Language universals and linguistic typology. Oxford: Blackwell.
Croft, William. 1990. Typology and universals. Cambridge: Cambridge University Press.
DeLancey, Scott. 1981. An interpretation of split ergativity and related patterns. Language 57. 626-657.
Dobrovie-Sorin, Carmen. 1994. The Syntax of Romanian. Berlin & New York: Mouton de Gruyter.
Dobrovie-Sorin, Carmen. 1995. Syntactic configurations and reference: se/si in Romance. In Karen Zagona (ed.), Grammatical theory and Romance languages, 73-86. Amsterdam: Benjamins.
Fried, Mirjam. 1990. Reflexives as grammatical constructions: a case study in Czech. In Kira Hall, Jean-Pierre Koenig, Michael Meacham, Sondra Reinman & Laurel A. Sutton. (eds.), Proceedings of the sixteenth annual meeting of the Berkeley Linguistic Society, 127-139. Berkeley: Berkeley Linguistic Society.
Givon, Talmy. 1976. Topic, pronoun, and grammatical agreement. In Charles Li (ed.), Subject and topic, 149-188. New York: Academic Press.
Givon, Talmy. 1979. On understanding grammar. New York: Academic Press.
Givon, Talmy. 1982. Transitivity, topicality, and the Ute impersonal passive. Syntax and Semantics 15. 143-160.
Grimshaw, Jane. 1997. Projections, heads, and optimality. Linguistic Inquiry 28. 373-422.
Grimshaw, Jane. 2001. Optimal clitic positions and the lexicon in Romance clitic systems. In Geraldine Legendre, Sten Vikner & Jane Grimshaw (eds.), Optimality-theoretic syntax, 205-239. Cambridge: MIT Press.
Grimshaw, Jane & Vieri Samek-Lodovici. 1995. Optimal subjects. In Jill Beckman, Laura Walsh Dickey & Suzanne Urbanczyk (eds.), Papers in Optimality Theory (University of Massachusetts Occasional Papers in Linguistics 18), 589-605. Amherst: GSLA.
Hidalgo, Raquel. 1994. The pragmatics of de-transitive voice in Spanish: From passive to inverse? In Talmy Givon (ed.), Voice and inversion, 169-86. Amsterdam: John Benjamins.
Hopper, Paul & Sandra A. Thompson. 1981. Transitivity in grammar and discourse. Language 56. 251-299.
Jackendoff, Ray. 1987. The status of thematic relations in linguistic theory. Linguistic Inquiry 18. 369-411.
Jelinek, Eloise & Richard A. Demers. 1983. The agent hierarchy and voice in some Coast Salish languages. International Journal of American Linguistics 49. 167-185.
Kiparsky, Paul. 2001. Structural case in Finnish. Lingua 111. 315-376.
Legendre, Geraldine, William Raymond & Paul Smolensky. 1993. An optimality-theoretic typology of case and grammatical voice systems. In Joshua S. Guenter, Barbara A. Kaiser & Cheryl C. Zoll (eds.), Proceedings of the nineteenth annual meeting of the Berkeley Linguistic Society, 464-478. Berkeley, CA: Berkeley Linguistic Society.
Lepschy, Anna Laura & Guilio Lepschy. 1977. The Italian language today. London: Hutchinson.
Lepschy, Guilio. 1978. Saggi di linguistica italiana. Bologna: Il mulino.
Levin, Beth & Malka Rappaport-Hovav. 1995. Unaccusativity: At the syntax-lexical semantics interface. Cambridge: MIT Press.
Levin, Beth & Malka Rappaport-Hovav. 2005. Argument realization. Cambridge: Cambridge University Press.
Manacorda de Rosetti, Mabel V. 1969. La llamada "pasiva con se" en el sistema espanol. In Ana Maria Barrenechea & Mabel V. Manacorda de Rosetti (eds.), Estudios de gramatica estructural, 91-100. Buenos Aires: Editorial Paidos.
Newmeyer, Frederick. 2002. Optimality and functionality: A critique of functionally-based optimality-theoretic syntax. Natural language and Linguistic Theory 20. 43-80.
Perlmutter, David. 1978. Impersonal passives and the unaccusative hypothesis. In Proceedings of the fourth annual meeting of the Berkeley Linguistic Society, 157-189. Berkeley: Berkeley Linguistics Society.
Perlmutter, David. 1983. Personal vs. impersonal constructions. Natural Language and Linguistic Theory 1. 141-200.
Perlmutter, David & Paul Postal. 1984. Impersonal passives and some relational laws. In David Perlmutter & Carol Rosen (eds.), Studies in Relational Grammar II, 126-170. Chicago: University of Chicago Press.
Prince, Alan & Paul Smolensky. 1993. Optimality theory: Constraint interaction in generative grammar (RuCCS Technical Report #2). Piscataway, NJ: Rutgers University Center for Cognitive Science.
Rivero, Maria Luisa. 2000. On impersonal reflexives in Romance and Slavic and semantic variation. In Joaquim Camps & Caroline Wiltshire (eds.), Romance syntax, semantics, and L2 acquisition, 169-195. Amsterdam: John Benjamins.
Sells, Peter. 2001. Form and function in the typology of grammatical voice systems. In Geraldine Legendre, Jane Grimshaw & Sten Vikner (eds.), Optimality-theoretic syntax, 355391. Cambridge, MA: MIT Press.
Siewierska, Anna. 1984. The passive: A comparative linguistic analysis. London: Croom Helm.
Siewierska, Anna. 1988. The passive in Slavic. In Masayoshi Shibatani (ed.), Passive and voice, 243-289. Amsterdam: John Benjamins.
Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W. Dixon (ed.), Grammatical categories in Australian languages, 112-171. Canberra: Australian Institute of Aboriginal Studies.
Sobin, Nicholas. 1985. Case assignment in Ukrainian morphological passive constructions. Linguistic Inquiry 16. 649-662.
Stiebels, Barbara. 2000. Linker inventories, linking splits and lexical economy. In Barbara
Stiebels & Dieter Wunderlich (eds.), Lexicon in focus, 211-245. Berlin: Akademie Verlag. Timberlake, Alan. 1982. The impersonal passive in Lithuanian. In Monica Maccaulay, Orin D. Gensler, Claudia Brugman, Inese Civkulis, Amy Dahlstrom, Katherine Krile & Rob Sturm (eds.), Proceedings of the eighth annual meeting of the Berkeley Linguistics Society, 508-524. Berkeley: Berkeley Linguistics Society.
Woolford, Ellen. 2001. Case patterns. In Geraldine Legendre, Sten Vikner & Jane Grimshaw (eds.), Optimality-theoretic syntax, 509-543. Cambridge, MA: MIT Press.
Wunderlich, Dieter. 1997. Cause and the structure of verbs. Linguistic Inquiry 28. 27-68.
Wunderlich, Dieter & Renate Lakamper. 2001. On the interaction of structural and semantic case. Lingua 111. 277-418.
(1.) The ideas in this article matured over a long period of time, so the list of people I should than is rather long. I benefited greatly from comments made by the audiences at the University of Texas Austin, Stanford University, the University of California, Berkeley, and Johns Hopkins University. I give special thanks to Geraldine Legendre, Paul Smolensky, Luigi Burzio, Joan Bresnan, and Orhan Orgun. All errors are my responsibility. Correspondence address: Department of Linguistics, University of California, Davis, One Shields Avenue, Davis, CA 95616, USA. E-mail: raranovich@ ucdavis.edu.
(2.) The arbitrary subject use of the reflexive can also be found in Quechua, Uto-Aztecan, some Turkic languages, (Turkish, Tatar, Uighur), some Australian languages (Lardil, Dja:bugay, Ngarinjin), and in African and Dravidian languages. Arbitrary human subjects also occur in periphrastic passives (Perlmutter 1978, 1983; Perlmutter and Postal 1984; Timberlake 1982; Sobin 1985; Givon 1982).
(3.) Siewierska (1984) notices a crosslinguistic tendency to exclude animate and pronominal patients from the subject of reflexive passives. In this article I make a related but complementary observation: inanimate patients are disallowed as objects of reflexive passives. My claim about animacy and the arbitrary se construction, then, is stronger than Siewierska's, since I suggest that animacy motivates an alternation in the realization of the patient.
(4.) The controversy about the markedness of objects with respect to animacy is summarized and documented in Croft (1990), primarily with respect to a conflict between case marking and agreement. He notices that "direct objects that are high on the animacy and/or definiteness hierarchies are case-marked, and therefore marked; but the same direct objects trigger agreement, which should indicate unmarked status" (Croft 1990: 160). He suggests that the conflict can be resolved by developing two different functional accounts, one for agreement and another one for case marking. In this article I focus my attention on the conflicting proposals about animacy and argument realization, a controversy that is not as clearly defined in the general literature.
(5.) The following abbreviations are used in the glosses throughout the article: ACC accusative, F feminine, M masculine, NOM nominative, PL plural, SG singular. The reflexive clitic that is used in arbitrary se constructions is glossed as SE.
(6.) The literature on arbitrary se offers wide support for the claim that the reflexive passive is different from the impersonal reflexive (Manacorda de Rosetti 1969; Lepschy and Lepschy 1977; Lepschy 1978; Rivero 2000). Besides the grammatical differences between the two constructions, evidence for the claim that they are distinct from each other is provided by their contrasting selectional and distributional requirements, and by their uneven crosslinguistic distribution. The reflexive passive, for instance, is unavailable for verbs that have no external thematic role, such as unaccusatives (Perlmutter 1978, 1983) or passivized verbs. Cinque's hypothesis that argument se absorbs an external thematic role accounts for this. This fact is observed in languages like Romanian and Czech, which have no impersonal reflexive constructions (Fried 1990; Dobrovie-Sorin 1994, 1995). But even in languages like Spanish or Italian, Cinque notices, the distribution of argument se and nonargument se is asymmetrical. Impersonal reflexives are excluded from nonfinite contexts. Nonargumental se is only able to combine with personal AGR (unavailable in nonfinite clauses) because the empty subject pronominal bears a thematic role, and needs to have f-features.
(7.) OT is not a theory of grammar (or syntax), but rather a framework in which different theories can be formalized. In the version of OT syntax presented in Grimshaw (1997), the input consists of a predicate and its arguments (in addition to relevant features for informational prominence of the arguments, etc.), and the candidates are alternative structural realizations of the input. Grimshaw assumes a model in which syntactic structure is generated by universal rules in GEN which project syntactic structure ac cording to the principles of X-bar theory, and that move heads and phrases around, leaving traces behind. Bresnan (2000) develops a conceptually similar approach based on the principles of lexical-functional grammar. Instead of movement rules, Bresnan uses correspondences between two levels of representation: c-structure (constituent structure) and f-structure (functional structure). The input is f-structure, and the candidates are alternative c-structures in incomplete correspondence with those f-structures. In both models, however, the input is unstructured lexical semantic content, and the candidates are alternative structural realizations of that content. Constraints may determine the structural position of a particular argument or predicate (e.g., whether the subject follows or precedes a verb), or the complexity of a clause (e.g., whether an auxiliary verb must be in the clause or not).
(8.) Aissen derives the constraint * Su/x by harmonic alignment of the scale Su > Oj and the scale of thematic prominence X > x, where X indicates a more thematic constituent than x.
(9.) Levin and Rappaport-Hovav's mapping theory shares many of the syntactic assumptions of some multistratal generative theories. In particular, patient-like arguments are always linked to a position internal to the VP. In cases where a patient surfaces as a subject, a syntactic rule that moves the argument out of its VP internal position is assumed. In these models, then, the problems presented by passives or by unaccusative verbs (i.e., intransitive verbs with patient subjects) do not arise, since the marked configuration in which a patient is realized as a subject only arises after arguments are projected onto the syntax.
(10.) The default rule assigns a [-r] feature to the most prominent argument, and a [+r] feature to all other underspecified arguments. This aspect of the LMT captures some of the insights behind the PPH, since the linking algorithm requires a minimal hierarchical ordering of the arguments, in which one of them outranks the rest. This minimal ordering may be guaranteed by a hierarchy of semantic roles, or by other means, such as predicate decomposition (Jackendoff 1987).
(11.) Feature decomposition theories of argument realization are also presented in Kiparsky (2001) and Wunderlich and Lakamper (2001). These authors also develop their theories in an optimality-theoretic framework, but focus on case assignment rather than grammatical functions. Some of their constraints, however, are similar to the ones I propose. My constraint UNIQ, for instance, is analogous to Wunderlich and Lakamper's "UNIQUENESS", a constraint that penalizes multiple occurrences of the same case in a clause.
(12.) Defaults are also used in Levin and Rappaport-Hovav's (1995) linking theory. Any arguments not linked to a syntactic position by the different linking rules they propose is projected as an internal argument by default. In an OT modeling of their theory, this default rule would also correspond to a low-ranking constraint against projecting arguments outside the VP.
(13.) This is consistent with the bound character of reflexive clitics, and with the reduced number of functions they can perform and meanings they can express. It is also consistent with research on their morphosyntactic properties. Grimshaw (2001), building on work by Bonet (1995), has argued that the combinatory properties of the reflexive clitic in clitic clusters in Romance follows from the fact that they are underspecified for all morphosyntactic categories (case, number, gender, etc.) except reflexivity.
(14.) Siewierska (1984: 165) gives a similar example as grammatical, but I disagree with her judgment.
(15.) These data underscore the need for further study regarding the status of inanimate and nonhuman proper nouns in the animacy hierarchy. Comrie (1981:189) cites the case of Chukchi reindeer names, which have the same grammatical properties as human proper nouns and are therefore higher up in the animacy scale than human common nouns. Other proper nouns, however, may not be as prominent.
(16.) In some languages pronouns are lexically or morphologically marked for animacy distinctions, but not in Spanish, where the contrasts are only in gender (the vestigial neuter demonstrative pronouns eso, esto, aquello, and the neuter definite article lo, are also gender markers). An open question is whether pronouns marked as inanimate will exhibit a reduced degree of individuation in languages that do so. An alternative approach to the one I have presented is to use local conjunction of * [+o]/inan and a constraint like * [+o]/common to make sure that inanimate common nouns are the most marked object types. Space limitations prevent me from discussing the issue further.
(17.) Notice that I am not claiming that Aissen's (1999) constraints are relational: they are not, and they should be replaced with relational constraints that account for the right generalizations. However, it is interesting to recall that Aissen acknowledges the relational nature of person hierarchy effects in her own analysis of inverse marking in Nocte. She derives relational constraints from the constraints in (19a) and (19b) through local conjunction. My suggestion is that relational animacy constraints should be stated as primitives.
(18.) An anonymous reviewer suggests that this approach may be extended to account for the alternation between the reflexive passive and the impersonat reflexive in Spanish: If accusative arguments incur less severe violations of the constraint set when they are higher in animacy, then assigning nominative to patients lower in the animacy scale is the optimal outcome. Unlike LFG's lexical mapping theory, and the OT implementation of it I have proposed, Stiebels' constraints are built on the basis of a direct association between semantic roles and case features, eschewing grammatical function. In LFG, on the other hand, grammatical function (specified at the level of f-structure) are central in the formulation of universal linguistic principles. This fundamental difference between the two theories makes it difficult to establish a direct comparison between the two approaches.
Table 1. Parameters of transitivity High transitivity Low transitivity A. Participants 2 or more (A and O) l participant B. Kinesis action nonaction C. Aspect telic atelic D. Punctuality punctual nonpunctual E. Volitionality volitional nonvolitional F. Affirmation affirmative negative G. Mode realis irrealis H. Agency A high in potency A low in potency I. Affectedness of O O totally affected O not affected J. Individuation of O O highly individuated O nonindividuated Table 2. Parameters of individuation Individuated Nonindividuated proper common human, animate inanimate concrete abstract singular plural count mass referential, definite nonreferential Table 3. Voice/animacy alternations in Salish (Lummi) Ag/3-Pat/3 Ag/3-Pat/Loc Active OK * Passive OK OK
|Printer friendly Cite/link Email Feedback|
|Publication:||Linguistics: an interdisciplinary journal of the language sciences|
|Date:||May 1, 2009|
|Previous Article:||Resultative constructions with "implied-result" and "entailed-result" verbs in Thai and English: a contrastive study.|
|Next Article:||Clausal coordination and coordinate ellipsis in a model of the speaker.|