The development of alignment systems
A grammatical domain that has been regarded as evidence for various types of psychological mechanisms is that of alignment systems, that is, the various ways in which the two arguments of transitive verbs and the only argument of intransitive verbs (henceforth, following the notation introduced in Dixon (1994), A, O, and S arguments) can be grouped together for grammatical purposes in the world's languages.
For example, different alignment systems conform to a general pattern whereby S arguments are usually encoded in the same way as either A or O arguments, while the latter two are usually distinguished, either by means of zero vs. overt marking or by means of different markers. This has been argued to reflect an ultimately economic psychological mechanism that leads speakers to distinguish between different argument roles only when they cannot avoid doing so, that is, when these roles could be confused. A and O arguments are encoded differently because they cooccur in transitive clauses, and could therefore be confused, while no separate encoding is needed for S arguments because they occur in isolation and cannot be confused with other arguments (Comrie 1978 and 1989), Dixon (1979 and 1994). This view is generally shared by functionally oriented typologists (see, e.g., Song 2001: 156-7 or Bickel 2011: 412), and it has also made its way into more formally oriented approaches such as Optimality Theory (see, e.g., de Hoop & Narasimhan 2005).
Another major type of explanation proposed for the cross-linguistic organization of alignment systems is that certain argument roles are encoded in the same way because they share particular properties. For example, A and S arguments may be encoded in the same way because they correspond to agentive participants, topical participants, or, more generally, participants that represent a starting point in discourse. S and O arguments may be encoded in the same way because they typically correspond to participants introduced for the first time in discourse, because certain types of S arguments correspond to nonagentive participants, or, in some analyses, because the participants most immediately involved in the state of affairs being described occur in S or O role. This type of explanation has been proposed in various versions in order to account for different alignment systems both within the functional-typological approach (Moravcsik (1978), Dixon (1979) and (1994), DeLancey (1981), Du Bois (1985), Mithun (1991), Mithun & Chafe (1999), Givon (2001), Song (2001), among several others) and in Cognitive Grammar (Langacker 1991: chap. 9).
Proponents of these explanations usually do not address the issue of how the shared properties of different argument roles may actually give rise to the various alignment patterns. There are, however, two possible mechanisms whereby this could take place, both of which are psychologically based and have been described in the literature as major mechanisms involved in language change (see e.g. Heine et al. (1991), Bybee et al. (1994), Hopper & Traugott (2003), Heine (2003), and Traugott & Dasher (2005)). The first is a mechanism of metonymization, or form-meaning redistribution in particular contexts. Constructions used to express some particular meaning may be associated over time with a cooccurring meaning. Thus, since A and S roles typically correspond to topical participants, constructions used to encode topics will be recurrently associated with these roles, and could over time be reinterpreted as expressing these roles. The second mechanism is one whereby speakers perceive a similarity between different concepts, and this leads them to express these concepts in the same way. Speakers could perceive a similarity between A arguments and certain types of S arguments because both involve a relatively high degree of agentivity, and this could lead them to use the same construction for the two (or extend individual constructions from one argument type to another).
A relatively large body of diachronic data is now available, however, on the development of different alignment systems in individual languages. In what follows, a selection of these data will be presented, and it will be argued that in many cases they do not actually provide any direct evidence for the psychological mechanisms that have been postulated in order to account for the relevant systems on synchronic grounds. In particular, the same system may originate from different processes in different languages, which are not obviously amenable to a single explanation and are independent of the need to distinguish between different argument roles, or the fact that particular roles share some properties. While this point has sometimes been raised for ergative systems (see, for example, Gildea (1998)), it can be generalized to all of the other major systems attested in the world's languages.
New markers for O and A arguments may originate through the grammaticalization of elements originally not used for the expression of argument roles. For example, new markers for O arguments have been shown to develop from 'take' verbs in constructions of the type 'take X and Verb (X)', which are reinterpreted as 'X OBJ Verb. Since the taking event is preliminary to the event described by the other verb, it may be less prominent in some contexts, at which point the 'take' meaning is obliterated and the verb is reinterpreted as merely indicating the O role that the entity being taken plays both in the taking event and in the event described by the other verb. This process has taken place, for example, in a number of West African languages (Lord (1993)) and in Mandarin Chinese (Li & Thompson (1981), among others). Example (1) illustrates the development of the O marker de from a former 'take' verb in Twi.
(1) a. okom de me hunger take me
'Hunger takes me' (Lord (1993): 70) [from an earlier description of the language]
b. o-de afoa ce boha-m he-OBj sword put scabbard-inside
'He put the sword into the scabbard' (Lord (1993): 66)
(literally 'He took the sword (and) put (it) in the inside of the scabbard')
c. aivua de kannea ni ahuhuru ma asase sun obj light and warmth give earth
'The sun gives ligth and warmth to the earth' (Lord (1993): 66)
(literally 'The sun takes light and warmth (and) gives them to the earth')
In some languages, the markers used for A arguments can transparently be related to indexical elements, that is, demonstratives and (third person) pronouns. This is illustrated in example (2) for the Australian language Bagandji. McGregor (2008)) accounts for this pattern by assuming that indexicals are initially used in apposition to nouns occurring in A roles to emphasize that these nouns (exceptionally) encode new information. As a result, they are reanalyzed as marking the A role.
(2) yadu-duru gandi-d-uru-ana
'This wind will carry it along / The wind will carry it along'(Hercus (1982): 63)
Another path to the development of A markers has been described by Rude (1991, 1997) for Sahaptian languages (Sahaptin and Nez Perce). In these languages, as can be seen from (3a-b), a Proto-Sahaptian suffix *-im (possibly derived from a verb 'come' in a serial verb construction) gave rise to a directional marker used on both verbs and nouns to indicate motion of an entity towards the speaker or the hearer. When attached to the A argument of a transitive clause, as in (3c), the suffix was reinterpreted as a marker of the A role. Due to its original semantics, its use in this function was initially restricted to clauses with third person A arguments and first or second person O arguments, though in Nez Perce, as can be seen from (4), it was later extended to all cases of transitive clauses with third person A arguments.
(3) a. aw i-q'inum-im-a wins
now 3NOM-see/look-csL-PAST man
'Now the man looked this way (Rude (1991): 41)
b. aw-nas i-q'inun-a wins-nim now-1sG 3NOM-see/look-PAST man-ERG
'Now the man looked at me' (Rude (1991): 41)
c. aw-nas xwisaat-nim i-twana-m-as now-1sG old.man-ERG 3NOM-follow-csL-iMPFV
'Now the old man is following me' (Rude (1991): 41)
Nez Perce (Penutian)
(4) wewukiye-ne pee-wi-ye haama-nm elk-OBj 3/3-shoot-PAST man-ERG
'The man shot an elk' (Rude (1991): 25)
These cases show that markers used for A and O arguments may develop from a variety of sources. All of the relevant processes, however, reflect a mechanism whereby particular aspects of the meaning of a complex expression (such as the description of a taking event, the signaling of new information, or that of directionality) are deactivated, and the forms originally associated with these meanings are reinterpreted as indicating particular argument roles that are relevant to the context. This process is crucially based on the contextual relationship between the original meaning of the forms and the argument roles they come to encode, for example, the fact that a 'take' verb in a serial verb construction has an O argument coreferential with the O argument of another verb, or the fact that indexical markers for new information or directional markers are used for A arguments. In this respect, these cases provide no evidence that the relevant markers arise out of the need to distinguish between A and O markers (although this does not rule out that this may be a factor in a speaker's synchronic use of these markers: more on this below).
Sometimes, different reinterpretation processes take place in a single language, leading to the development of distinct markers for different roles. In this case, the resulting system may be one where A, S, and O arguments are all encoded differently. This was the case in Sahaptian, where, in addition to the development of new A markers, a separate process led to the reinterpretation of a Proto-Sahaptian directional marker * -(n)en 'thither' (possibly related to a verb 'go') as an allative, dative, and eventually accusative marker (illustrated in (4) above). This led to a system with zero marking for S arguments and different overt markers for A and O arguments (Rude 1991 and 1997). Often, however, processes leading to the development of new markers for particular argument roles take place in languages where all roles are initially encoded in the same way (the majority of the languages taken into account in the cross-linguistic survey of alignment of case marking of full NPs presented in Comrie (2008)). In this case, depending on whether the new marker is used for A or O, the process may yield patterns where S and O are undifferentiated and A is encoded differently, or patterns where A and S are undifferentiated and O is encoded differently, that is, ergative or nominative patterns. This, however, is unrelated to any similarity between different roles that is reflected at the linguistic level. Rather, since all roles were encoded in the same way in the original system, the roles that are not involved in the process of reinterpretation remain undifferentiated in the resulting system.
Another major process leading to the development of new alignment systems is the reanalysis of the argument structure of particular constructions. In this case, the properties of the new system are a direct result of the argument encoding pattern of the source construction. A well-known instance of this process is the development of ergative systems from passive constructions. Over time, constructions of the type 'X is done by Y' may become functionally equivalent to their active counterpart 'Y does X'. In the resulting transitive clause, X, the O argument, has the same marking of the S argument of the passive construction from which it is derived, while Y, the A argument, is derived from the agent of the passive clause, and maintains the oblique marking of the latter. This process has been postulated, for example, for the Hindi perfective construction in (5), whose presumed Sanskrit antecedent is reported in (6) (see Verbeke & De Cuypere (2009) for a recent comprehensive discussion of the issues involved).
Hindi (Indo-European) (5) lark-e=ne bacch-e=ko m~ar-a hai
boy-OBL-ERG child-OBL-ACC hit-PERF.M.SG be.AUX
'The boy has hit the child' (Verbeke & De Cuypere (2009): 5)
(6) devadatt-ena kata-h kr-tah Devadatta-INSTR mat-NOM make-NOM.PAST.PTCPL
'The mat is made by Devadatta' (Verbeke & De Cuypere (2009): 3)
Gildea (1998) provides a detailed analysis of similar processes of reanalysis in Cariban languages. In this case, the process starts from complex sentences involving nominalized verb forms, which may give rise to either nominative or ergative systems as they are reinterpreted as monoclausal structures. For example, Gildea argues, the Wayana progressive construction illustrated in (7) originated from constructions of the type 'X is occupied with Verbing', or 'X is occupied with the Verbing of Y', which were reinterpreted as intransitive or transitive monoclausal structures, that is, 'X is Verbing' or 'X is Verbing Y'. This yielded a nominative system, in that, in the resulting constructions, A and S arguments are encoded in the same way because they both originate from the S argument of the main clause in the source construction, while O arguments are encoded differently because they originate from the possessor argument of the nominalized verb in the source construction and maintain the original possessor marking. The Carina future construction illustrated in (8) originated from constructions such as 'It will be X's Verbing', or 'To X It will be Y's Verbing', in which the nominalized verb was possessed by its notional O or S argument, while its notional A argument was encoded as a dative. These too were reinterpreted as monoclausal structures, that is, 'X will Verb' or 'X will Verb Y'. This yielded an ergative pattern, in that, in the resulting constructions, S and O arguments originate from the possessor argument of the nominalized verb, and maintain possessor marking, while A arguments originate from the dative NP, and maintain dative marking.
(7) i-pakoro-n iri-0 pdk wai 1-house-POSS make-NOMLZR occ. with 1.be
'I'm (occupied with) making my house (lit. 'my house's making')
(Gildea (1998): 201)
(8) a. i-woona-ri-ma
'I will cultivate' (Gildea (1998): 169)
'(Somebody) will take me' (Gildea (1998): 169)
c. a-eena-ri-ma i-'wa
'I will have you' (Gildea (1998): 170)
Reanalysis of argument structure has also been argued to be the source of active systems (see, e.g. Harris (1985), Malchukov (2008) and Mithun (2008)). For example, transitive constructions with unexpressed third person A arguments, e.g. '(It) burned me', can be reinterpreted as intransitive ones, e.g. 'I am burned', where the verb describes the state resulting from the action described by the transitive clause. The O argument of the transitive verb becomes the S argument of the intransitive one, so that the two come to be encoded in the same way. This process has been postulated for several language families by Malchukov (2008) and Mithun (2008), and some languages provide direct diachronic evidence for it. For example, Holton (2008) shows that, in Galela, the formal identity between the S argument of stative intransitive verbs and the O arguments of transitive verbs, illustrated in (9a-b), originated from the fact that intransitive clauses with stative verbs were originally transitive clauses with a third person nonhuman A argument cross-referenced by a verbal prefix i-. In the late 19th century, as can be seen from (9c-d), this prefix became optional and eventually disappeared, which led to the reinterpretation of the transitive clause as a corresponding intransitive one.
Galela (Austronesian) (9) a. ni-kiolo
'You are asleep' (Modern Galela: (Holton (2008): 261))
b. wo-ni-doto 3M.SG.A-2SG.U-teach
'He teaches you' (Modern Galela: (Holton (2008): 261))
'She is angry' (19th century Galela: (Holton (2008): 272))
'She is old' (19th century Galela: (Holton (2008): 272))
In all of these cases, the way in which individual arguments are encoded is a result of the way in which the corresponding NPs are encoded in the source construction. Thus, for example, A and O have dedicated marking because they inherit the oblique or possessor marking of the corresponding NPs in the source construction. Likewise, different argument roles are encoded in the same way because one role is reinterpreted as another and maintains its original marking, or because they had the same marking in the source construction. In this respect, the relevant alignment systems do not obviously originate from the need to distinguish between A and O arguments, or any perceived similarity between different argument roles. Rather, what gives rise to these systems are whatever factors motivate the re-interpretation of the source construction, for example the fact that passive clauses describe the same semantic situation as the corresponding active ones, the fact that certain types of complex sentences describe the same situation as the corresponding monoclausal structures, or the processes of inference that lead speakers to reinterpret transitive clauses as signaling a state resulting from an action rather than the action itself.
All this does not rule out that, in some cases, particular alignment patterns may originate from the need to distinguish between A and O arguments, or from some shared property of different roles that is specifically encoded in the pattern. For example, König (2008: 273-6) shows that, in !Xun (Khoisan), a topic marker is in the process of developing into a marker for A and S arguments. This is naturally accounted for by assuming that the topic marker is recurrently associated with A and S arguments because they are usually topical, and this has led to its being reinterpreted as a marker for these arguments. Likewise, some languages use different marking for A and O arguments only when these are semantically similar, for example, when they are both animate, which might be evidence that, in this case, the marking is indeed being used to distinguish between these two roles (Comrie (1978), Bickel (2011), among others). The point is, however, that the data discussed in this section, while by no means exhaustive of the range of processes that may give rise to alignment systems, suggest that the latter are basically an epiphenomenal result of different diachronic processes in different languages and constructions (for a recent study in a similar vein, see Creissels (2008)). It follows that any assumption about the psychological mechanisms underlying particular alignment systems should be tested against the diachronic processes that give rise to these patterns. These processes may uncover psychological mechanisms insofar as they point to a number of principles that lead speakers to recombine the formal and the meaning components of a complex expression. These mechanisms, however, may or may not be the same as those that can be postulated on synchronic grounds, and they account for the evolution of specific constructions, rather than the relevant alignment systems in general.
-  A related question is, of course, what motivates the alignment pattern used in the source construction in the first place, for example the use of the same marking for S and O, or the use of distinct marking for A and O. This, however, has to be investigated separately for each construction, and it is quite possible that, in this case too, individual alignment patterns cannot obviously be related to the need to distinguish between different argument roles, or the fact that some roles share some properties. For example, as far as nominalizations are concerned, it has been observed that the use of possessor marking for both S and O arguments might be unrelated to any perceived similarity between these arguments in themselves. Rather, this use might originate from the fact that S and O arguments are usually the only overt arguments in a nominalization, and they independently receive possessor marking in analogy with non-derived NPs (Koptjevskaja-Tamm (1993b): 260, Gildea (1998): 122-3).
-  Another process that could be based on some shared property of A and S arguments is the extension of individual markers from the former to the latter. This process has been postulated for some languages (see, e.g., König (2008) on several African languages, and Li et al. (1977) on Wappo), but the relevant literature often does not provide any data on the etymology of the relevant markers or the various steps of the process, so it is unclear what properties exactly could be involved in these particular cases.
-  This fits in with the fact that, as repeatedly pointed out in the literature (Dixon (1994), Kazen-in (1994), Croft (2001), Siewierska (2004), Bickel (2011)), the synchronic distribution of alignment sys- tems is construction-specific, that is, individual alignment systems are typically limited to particular constructions, rather than characterizing the grammatical organization of the language as a whole.