Evidence against unidirectionality in the emergence of middle voice systems

3.1 The functions of the Hittite middle voice

Studies on the Hittite middle voice usually set up a distinction between two main classes of Middle verbs: media tantum and alternating verbs (Friedrich 1960; Hoffner and Melchert 2008: 303–305; Inglese 2020; Luraghi 2012; Neu 1968a). This means that the Hittite middle voice effectively functions as a MM as defined in Section 2.

The group of media tantum correspond to what we call non-oppositional verbs. These are verbs that only inflect in the middle voice throughout the history of the language. Examples of media tantum are given in (2) (for an exhaustive list see Inglese 2020: 108–114):

As shown in (2), Hittite media tantum cover situation types that are typical of middle marking (as per Kemmer 1993), including verbs of position and posture, e.g. ar- ^ta(ri) ‘stand’ and es- ^a(ri) ‘sit down’, verbs of motion, e.g. ye/a- ^tta(ri) ‘go’, and spontaneous events, e.g. ur- ^āri ‘burn’. Generalizing over the semantics of the media tantum, a clear pattern emerges whereby they encode either states, e.g. ki- ^tta(ri) ‘lie’, or, more frequently, uncontrolled change-of-state events, e.g. kīs- ^a(ri) ‘become’ (Inglese 2020: 119–122; Luraghi 2012; contra Neu 1968a: 117–119). Syntactically, most media tantum are intransitive verbs of the unaccusative type, defined in Hittite as a specific subset of intransitive verbs that require 3rd person enclitic subject pronouns in the absence of an overt subject (Garrett 1996).^[5] In this respect, they differ from both transitive and intransitive unergative verbs, which do not allow enclitic subjects under any circumstance. Consider example (3) in which the Middle verb es- ‘sit down’ is found with the subject clitic =as.

With verbs that inflect both in the Active and in the middle voice, voice alternation encodes a number of valency-reducing operations (Hoffner and Melchert 2008: 303–305; Inglese 2020: 131–151): reflexive, reciprocal, passive, and anticausative (as defined in Zúñiga and Kittilä 2019). In the first place, oppositional Middle verbs may have a reflexive and a reciprocal interpretation. In (4), the Middle form suppiahhut ‘purify oneself’ is the reflexive counterpart of the transitive active verb suppiahh- ⁱ ‘purify’, while in (5), the plural form zahhanda is a reciprocal verb based on Active transitive zahh- ‘hit’.

Another, comparatively frequent, function of the middle voice is passive. In the passage in (6), the verb handāittari ‘be determined’ is the passive counterpart of transitive active handai- ‘determine (through oracle)’.

In (6), a passive interpretation is confirmed by the occurrence of the Agent phrase IŠTU ^dUTU-ŠI ‘by his Majesty’. Note further that in this example, the verb handāittari triggers the use of the subject enclitic pronoun =at, which demonstrates that oppositional Middle verbs are syntactically intransitive unaccusatives, much in the same way as intransitive media tantum like es- in (3). This proves that the middle voice effectively operates as a valency-reducing device in Hittite.

Finally, as also discussed by Luraghi (2012), voice alternation can also be associated with anticausativization. Consider example (7).

In (7), the Middle form zinnattat encodes a spontaneous event ‘come to an end’ (this being the only possible interpretation, given the nature of the subject gimmanza ‘winter’) as opposed to the transitive Active counterpart zinni- ^zi ‘complete, finish (tr.)’, which indicates an externally induced event. Notably, middle forms of several verbs are ambiguous between an anticausative and a passive interpretation, e.g. Active tarupp- ^zi ‘gather (tr.)’ versus Middle tarupp- ^ta(ri) ‘gather (intr.), be gathered’.^[7]

3.2 The chronological distribution of the Hittite middle

From a panchronic perspective, i.e. if one takes into consideration the entire Hittite corpus, oppositional Middles are more frequent than non-oppositional ones, as reported in Table 2.^[8] At first sight, this would seem to suggest that the oppositional component of the system is diachronically primary. Zooming in on oppositional verbs, as the data in Table 3 shows, the passive function is the most widespread, closely followed by the anticausative, while occurrences of reflexives and reciprocals are quite marginal.

Things become much more interesting once the data is put into a chronological perspective. Thanks to advances in techniques of paleographic text dating, we can now rely on a careful distinction of Hittite texts into three distinct language stages: Old, Middle, and New Hittite (see Inglese 2020: 62–63; van den Hout 2009). This means that, with respect to earlier studies such as Neu (1968a, 1968b), the chronological distribution of Middle verbs can now be explored in detail over the entire history of the language.

Let us first consider the distribution of the two macro-classes of Middle verbs. Overall, as shown in Table 2, non-oppositional verbs are slightly outnumbered by oppositional Middles. The chronological distribution offers a radically different picture, however, as suggested by a comparison between Tables 2 and 4.^[9] The most striking result is that the distribution of the two classes is not even over time. In OH, most Middles belong to the non-oppositional class, while oppositional verbs have a rather limited distribution. This imbalance is also reflected in token frequency at this stage: non-oppositional Middles amount to 191 tokens in the OH corpus, versus 24 tokens of oppositional verbs. The OH distribution is radically altered over time, leading to the NH situation, which can be considered the mirror image of OH. At this stage, non-oppositional Middles constitute the minority, while oppositional verbs show signs of a strong expansion.

The internal distribution of oppositional Middles also varies over time. Table 3 suggests that passive is the most frequent function of oppositional Middle verbs in the entire corpus. However, a closer look at the chronology of oppositional Middles, as in Table 5, shows that passives are relatively limited in OH, and become increasingly common over time (anticausatives also increase over time but at a less pronounced rate).^[10] Reflexives and reciprocals remain rare at all stages.

3.3 The origin of the middle voice in Hittite

Taken at face value, the panchronic distribution of the functions of the Hittite middle voice is similar to that of Italian si discussed in (1). In both languages the MM applies to oppositional verbs with passive, anticausative, reflexive, and reciprocal functions as well as to a class of non-oppositional verbs. Based on this similarity, and keeping in mind the general validity of the reflexive > mm shift, one could speculate that the Hittite middle voice originated as a reflexive marker. In fact, a reflexive origin has sometimes been proposed for the Proto-Indo-European Middle inflection as a whole (e.g. Meier-Brügger 2010: 396).

However, the Hittite corpus does not fully support this reconstruction. To begin with, it must be mentioned that even synchronically, voice alternation is only weakly connected with reflexivity in Hittite. Instead, reflexivity is primarily encoded by means of the so-called Reflexive particle =za in combination with Active verbs (see Cotticelli Kurras and Rizza 2013; Hoffner and Melchert 2008: 357–360; Inglese 2020: 81–86). The reflexive usage of =za is exemplified in (8), where the particle combines with the otherwise transitive Active verb walh- ‘hit’ to yield a reflexive reading ‘hit oneself’.

Moreover, even the few oppositional Middles that license a reflexive reading tend to co-occur with =za in reflexive contexts (see Inglese 2020: 147–148). In fact, the example in (4), with the bare Middle suppihhut in reflexive function, is the exception, as the same verb suppiyahh- ‘purify’ also occurs with the particle when used reflexively, as in (9).

The low frequency of Middle verbs in reflexive function, combined with the existence of a more productive alternative for the encoding of reflexivity, makes it doubtful that reflexivity could have been the original semantic core of the Middle inflection. In addition, as remarked in Section 3.2, in the oldest stage of the language Middle verbs are predominantly non-oppositional, and oppositional ones gain ground only at a later stage. In other words, the Hittite corpus data hardly supports the oppositional > non-oppositional shift.

Based on these observations, the Hittite data can best be explained by assuming a different development, as already proposed by Luraghi (2012) and Cotticelli Kurras and Rizza (2013), among others. The OH data presented in Table 4, in which most middle verbs belong to the media tantum, supports the hypothesis that verbal voice might have been lexically determined in origin, with verbs only inflecting either in the Active or in the middle voice (this idea is not entirely new in IE studies; see e.g. Lazzeroni 2004: 142). In other words, the Middle inflection was originally confined to non-oppositional Middles, thus suggesting that oppositional Middles are an innovation.

As already noted, a remarkable feature of media tantum in OH is that they are mostly connected with spontaneous change-of-state events involving inanimate subjects, e.g. kis- ‘happen’, kist- ‘go out (of fire)’ (see Inglese 2020: 124). Spontaneous Middles of this sort could have easily been perceived as being in opposition to Active verbs indicating the corresponding externally caused event. Indeed, a pattern of voice opposition contrasting Active and Middle verbs already exists between intransitive non-oppositional Middles and their transitive causative counterparts built with derivational suffixes such as -nu- and Active endings, e.g. kist- ^ā(ri) ‘go out’ versus kistanu- ^zi ‘put out, extinguish’ (on this pattern see Luraghi 2012). Similarly, Active versus Middle pairs also include suppletive pairs of the type kis- ^a(ri) ‘happen.mid’ versus iya- ^zi ‘cause to happen.act’. The existence of this polarization between the Active and the middle voice in the lexicon could have led to the use of the same verbal root in both voices to indicate the anticausative alternation. This might be how voice alternation arose in the first place in the language, making it reasonable to assume that the anticausative function was the first oppositional usage of the middle voice that emerged from the original lexical distribution.

Once established as an anticausativization strategy, voice alternation could have spread to other oppositional functions as well (see Inglese 2020: 228–241 for a full account). The shift from anticausative to passive does not need much comment, as it is typologically rather trivial (see e.g. Bahrt 2021: 206–207; Haspelmath 1990; Inglese 2020: 231–234). Links towards reflexives and reciprocals are less straightforward, and can be explained as follows (see also Bahrt 2021: 203–206).

The shift anticausative > reflexive/reciprocal can be better understood by adopting a finer-grained distinction of anticausatives in terms of decausatives and autocausatives (Creissels 2006: 10; Geniušienė 1987: 86–89, 98–104). Decausative verbs such as ‘melt (intr.)’ typically involve non-controlling Patient subjects, while the subject of autocausatives can be construed as partly exerting control on the onset of the change-of-state event. Autocausatives typically include verbs of self-induced motion, e.g. ‘turn’, which offer a possible bridging context between the anticausative and the reflexive domain.

Verbs of motion arguably offer the key bridging contexts from the anticausative to the reflexive function (a similar point is also made by Holvoet 2020: 118–119). At the pre-Hittite stage, as we have noted, voice alternation was likely restricted to verbs expressing a spontaneous change of state, i.e., decausatives. While verbs denoting a physical change of state (e.g. ‘melt’) constitute the core of the category, decausative verbs also include verbs of motion with inanimate subjects, on the ground that change of location also entails a change of state. Crucially, however, motion events differ from prototypical decausatives (physical change-of-state events), in that their single participant shares properties of prototypical Agents, i.e. agenthood and control, and Patients, i.e. undergoing a change of location (e.g. Luraghi 2010: 140). In other words, unlike events of physical change, motion events can more easily be construed as featuring either a non-controlling (decausative) or controlling (autocausative) subject. It is the autocausative use with motion verbs that represents the first step towards the reflexive use. Consider the behavior of oppositional Middle forms of the Hittite verb nini(n)k- ^zi ‘raise, set in motion’ in (10).

In (10a), the Middle form ninikanta occurs with the inanimate subject karittes ‘flood’ and the entire situation is depicted as a spontaneous change-of-state event. By contrast, in (10b) the very same verb is used with an animate subject ^LÚKÚR ‘enemy’, which is construed as agentively bringing about the motion event. Starting from autocausative contexts such as in (10b), voice alternation could have been further extended to reflexive contexts proper, leading to the final stage of the development, in which voice alternation applies to events that cannot possibly be construed as spontaneous and that exclusively involve animate controlling subjects, as in e.g. (4) above. A parallel cline of development can also be assumed for the reciprocal function, with autocausative verbs of the spatial lexical reciprocal type serving as bridging contexts, e.g. tarupp- ‘gather’ and sarra- ‘split, divide’ (see also Inglese 2017). The development of the oppositional functions of the Hittite Middle inflection can be summarized as in Figure 1.

Figure 1:

The development of oppositional functions of the Hittite middle voice (adapted from Inglese 2020: 240).

The development proposed here is also wholly compatible with the distribution of Middle verbs in the Hittite corpus, as discussed in Section 3.2. In the first place, assuming a non-oppositional origin of the Middle inflection best explains why non-oppositional Middles are more frequent in OH, and why oppositional ones become more frequent at later stages, as per Table 4. In particular, the fact that non-oppositional Middles seem to decrease over time is also due to the fact that some of these develop a historically new pattern of voice alternation, thereby entering the oppositional class. For example, the verb lazziye/a- ^tta(ri) ‘recover, become good’ is only non-oppositional in OH, but in NH, one also finds an Active oppositional counterpart (with causative meaning) lazziye/a- ^zi ‘set straight’.

By contrast, if one were to reconstruct an original reflexive function, one would have to assume that oppositional functions developed into non-oppositional verbs, which then became more prominent, only for oppositional functions to regain ground at a later stage. While this reconstruction is not per se impossible, it is unnecessarily complicated and does not rest on any independent evidence to postulate a reflexive origin of the Middle inflection.^[11]

In the second place, the development of oppositional functions sketched in Figure 1 is also compatible with the distribution of oppositional functions in the Hittite corpus as described in Table 5. Indeed, anticausatives are relatively common in OH already and remain so at later stages. The passive, which is the function most closely related to the anticausative, is a secondary development, as it becomes increasingly more frequent especially in NH. The extension to reflexive and reciprocal contexts, which, as discussed, follows a less mainstream path of development, never gained much ground, possibly on account of the existence of dedicated strategies for the encoding of these two functions (Inglese 2020: 156–163).

A typological peculiarity of the Hittite Middle also concerns the timing of the development of the reflexive function. Kemmer (1993: 229) argued that “middle markers from non-reflexive sources will not develop into markers of reflexive semantics”. However, Hittite offers evidence for precisely this type of unexpected development, with a middle marker originating in spontaneous/anticausative events extending to the reflexive domain only secondarily (for further parallels from unrelated languages, see Inglese 2023). It is also noteworthy that lack of directionality in the reflexive > mm development is also compatible with a general lack of directionality in the voice domain that has been advocated in more recent studies. In fact, as pointed out by Bahrt (2021: Chap. 7), it is reductive to claim that reflexives are unidirectionally the preferred source of other voice operations, as patterns of voice syncretism can in fact come about following different paths and taking different voice operations as input. Why it is specifically in the voice domain, and not in others, that we fail to see directionality in diachronic developments, is a question that we leave for future research (see Giomi and Inglese forthcoming for some preliminary remarks).

4.1 Impersonals as a middle voice system

The relevant structures are illustrated in (11) and (12) below, in contrast with the standard transitive structures in the language illustrated in (13) and (14). We will initially refer to them as ‘impersonal’ structures, in line with previous work (Verstraete 2011a), but their middle status will be demonstrated below.

Standard transitive structures in Umpithamu are characterized by a number of features, two of which are relevant in this context. First, they typically have cross-referencing pronouns, organized on a nominative-accusative basis. The system can be called ‘cross-referencing’ because it is not as systematic as full-fledged agreement (compare Siewierska 2004): Nominative pronouns are present in about 2/3 of verb-headed clauses (as in (13) and (14) above), while accusative pronouns are only present in about 1/3 of clauses (compare (13), without an accusative pronoun, with (14), which has one). Pronouns form complexes that are typically encliticized to the verb, as in (13) and (14), but they can also be located at the start of the clause, a characteristic to which we will return in Section 4.2 below. The second feature of standard transitive clauses concerns nominals, which are organized on an ergative-absolutive basis, as shown in (13) above for the ergative marker. Ergative marking is not obligatory in transitive clauses, but organized along principles of animacy and information structure, as is typical of so-called ‘optional ergative’ systems (Chappell and Verstraete 2019; McGregor 2010; Verstraete 2010). Inanimates take obligatory ergative case, while animates only take ergative case in specific information-structural contexts, like the focus context in (13), where the cause of a disease has just been established as a foreign object in the body, and the ergative-marked participant in (13) identifies the entity responsible for inserting it.

The impersonal structures in (11) and (12) use the same transitive verbs as in (13) and (14), but they differ from standard transitive structures in a few important respects. First, they never take nominative pronouns, as shown in (11) and (12). In other words, there is no trace of subject marking, although in some cases an Agent-type role can be discerned discursively, and more rarely expressed with a (non-cross-referenced) ergative-marked nominal, as in (15) below. Second, they take genitive pronouns instead of accusative ones for the Patient argument, and express these obligatorily rather than optionally. If such arguments are present in lexical form, they take absolutive case, again as illustrated in (15) with apa’ala ‘lower leg’, in an external possession relation with the Patient.^[12]

Morphosyntactically, the properties of the structures in (11), (12), and (15) can be interpreted as pointing to impersonal status, as argued in Verstraete (2011a), to which we refer for more details, since the argument for impersonal status is less directly relevant to the present discussion. What is more relevant, however, is the semantics of impersonal structures. The use of an oblique (‘genitive’) pronoun instead of the standard accusative one points to the presence of an Experiencer role: what all of these structures have in common is a feature of involuntary experience for a human Patient.^[13] This feature manifests itself semantically in two ways. In instances like (11), (15), and (16)–(17) below, impersonal structures use transitive ‘impact’ verbs to build idioms designating the experience of involuntary physical processes in human beings, like cramps (15), hiccups (16), sudden weaknesses (11), or pains (17).

In instances like (12) and (18) below, by contrast, impersonal structures provide an alternative construal for any transitive verb, where the basic semantics of the verb is maintained (unlike in the idioms above), but overlaid with a semantic layer that highlights the involuntary experience of the Patient in the process (and at the same time demotes the Agent). Structures like (12) and (18) are either used in contexts where a human being is inherently involuntary in experiencing an activity (as in (18), where an unconscious person is carried away), or in contexts where involuntary involvement is discursively highlighted (as in (12), which describes a government-ordered deportation away from people’s home country).

This double function of the structure is precisely what makes it a middle voice system from a typological perspective, in the sense used in this paper. It may not look like the canonical middle, since it does not use a dedicated inflectional marker (as in the case of Hittite in Section 3.1). Instead, we are dealing with a construction that can only be defined in terms of a specific configuration of features, viz. a transitive verb with obligatory absence of nominative cross-reference, and obligatory presence of genitive cross-reference for the Patient. But it is a middle system nonetheless, since it combines a non-oppositional distribution in the lexicon with a voice- or valency-related oppositional function in morphosyntax. The lexical function relates to the building of idioms for the very specific semantic domain of involuntary physical processes in human beings. The verbs involved are not themselves verbs of physical experience, but the constructions as a whole function as idioms for a range of sudden bodily processes over which people have no control. The morphosyntactic function relates to the use as an alternative construal for any transitive structure with a human Patient. Like most languages in the region, Umpithamu does not have any valency-changing morphology, but this structure is the functional equivalent of a more classic valency-reducing process like a passive. For one thing, structures like (12) and (18) demote the Agent: Ergative-marked nominals are rarely expressed, and even if they are, they are never cross-referenced by a nominative pronoun, which is the basic way to establish a subject-predicate relation in the language (compare 19a, which is a noun phrase, with 19b, which is a clause).

Secondly, the structure also highlights the Patient: cross-reference is obligatory (unlike with accusative pronouns), and the genitive case overlays it with Experiencer semantics, as being involuntarily involved in the activity. The main difference from a canonical passive is that this argument does not have any morphological subject characteristics, but there are, in fact, typological equivalents of such passives, where the promoted argument behaves like a subject in some ways, even if its morphological coding is not that of a subject (as shown in Malchukov 2008). In any case, the basic point is that this structure has all the properties of a middle voice system as defined in this paper: a restricted set of uses in the lexicon, in a set of idioms for a specific semantic domain, and a productive use in morphosyntax, as a passive-like structure that provides an alternative construal for any transitive structure with a human Patient.

4.2 Diachronic development

From the perspective of this paper, the Umpithamu structure is interesting in that we can reconstruct its historical development: This shows that the oppositional (valency-reducing) function developed from the non-oppositional (lexical) one rather than the other way round. The key in this argument is the system of cross-referencing pronouns that defines the middle voice system in Umpithamu, which can be shown to originate in contact with languages from the neighboring Lamalamic subgroup to the south. This is, in fact, just one instance of profound Lamalamic influence on the morphosyntax of Umpithamu, fueled by long-standing social links and patterns of multilingualism (see Rigsby 1997; Verstraete 2011a, 2012, 2018 for more examples and details of the contact situation).

The first step in the argument about origins concerns the basic morphosyntax of cross-referencing pronouns in Umpithamu. As already mentioned, pronouns invariably form complexes if there is more than one in a clause, and they can be found in two positions: Single pronouns and complexes are most typically encliticized to the verb, as in all of the examples discussed above, but they can also be found in clause-initial position in specific information-structural contexts, as in (20) below. What is peculiar is that the whole complex moves to initial position as soon as one of the participants is in focus or contrastive topic position, as is the case with the focused Agent participant in (20).

This particular morphosyntactic configuration is specific to Umpithamu within its subgroup. It is not found in any other Middle Paman language, and not in any other Australian language in fact, apart from the Lamalamic languages mentioned below (see Verstraete 2012 for evidence). There is a whole range of first-position phenomena affecting pronouns in Australia (see Mushin and Simpson 2008), but crucially not with the morphosyntactic specifics as found in Umpithamu. The only exception are the Lamalamic languages just south of Umpithamu, all three of which have exactly this configuration: Pronouns form complexes that can either be encliticized to the verb, or move to clause-initial position as soon as one of the participants referenced in the complex is in focus or contrastive topic. This is illustrated in (21) and (22) below for Umbuygamu and Lamalama, the two best-described languages in the subgroup.

In other words, this particular configuration appears to be an innovation within Lamalamic. Given how specific the configuration is, the fact that it is also found in Umpithamu just north of Lamalamic obviously suggests a contact scenario. The analysis in Verstraete (2012) provides linguistic and ethnographic evidence to show that the cross-reference pattern in Umpithamu does in fact originate in contact with Lamalamic languages. On the one hand, there are morphosyntactic peculiarities in Umpithamu that point to a relatively recent origin of the cross-referencing system: The most notable feature is the length of the pronoun complexes in enclitic position, which suggests inherited free pronouns that have been slotted into a borrowed pattern relatively recently. And on the other hand, there is good ethnographic evidence for patterns of intermarriage between members of Umpithamu and Lamalamic clans, leading to systematic multilingualism with Umpithamu and at least one Lamalamic language. The fine detail of the argument can be consulted in Verstraete (2012), but the basic point is that the morphosyntactic pattern of pronominal cross-reference in Umpithamu originates in Lamalamic, with the inherited Middle Paman pronoun material slotted into the pattern.

The second step in the argument concerns how these cross-referencing pronouns are used in the language. The impersonal configuration found in Umpithamu – in particular the systematic lack of nominative cross-reference and its associated Experiencer semantics – is not found in any of its Middle Paman relatives. Partially parallel structures known as ‘Experiencer object’ constructions are found further afield in Australia, mainly in some non-Pama-Nyungan families in the continent’s central north (see Evans 2004; Merlan 1985; Walsh 1987 for more details). In the immediate region, however, there is one almost exact parallel, once again in the Lamalamic languages. All three Lamalamic languages have a construction characterized by obligatory absence of nominative cross-reference, and obligatory presence of a non-nominative pronoun for the Patient, either an oblique form, like the dative in Rimanggudinhma in (23), or an accusative, as in the Umbuygamu structure in (24). Verstraete (2011a) provides more details on these structures.

In terms of semantics, the structures are an exact match for the non-oppositional (lexical) part of the Umpithamu structure: Again, they use basic transitive verbs in an impersonal structure, to build idioms for the domain of involuntary physical processes. In other words, it looks as if, along with the basic patterns of pronominal morphosyntax, more specific configurations like impersonal constructions were borrowed from Lamalamic into Umpithamu, as a very specific instance of pattern replication in verbal constructions (compare Grossman 2021). In most cases, even the transitive verbs used in the idioms match between the languages, e.g. ‘spear’ for hiccups, ‘pull’ for headaches or ‘bite’ for cramps.

Given this scenario about origins, we can now evaluate the diachronic development of the middle voice system in Umpithamu, which has both restricted non-oppositional uses in the lexicon and a productive oppositional use as a passive-like structure in morphosyntax. The equivalent structures in Lamalamic have the exact same lexical uses in building idioms for involuntary physical processes, but there is no trace of a productive passive-like use as in Umpithamu. This leads to the conclusion that only the lexical uses could have been carried over from Lamalamic to Umpithamu, and that the morphosyntactic function of Umpithamu’s middle voice system is new – or at least newer than its lexical function, and most likely developed out of this function. We cannot determine the trajectory with certainty (for instance, there might have been something else in the language that latched onto the incoming structure), but there is one small indication of how the development could have happened. There is a minor pattern, in Umpithamu and the Lamalamic languages, that resembles the impersonal construction in most ways, and that could have served as a bridge from the lexical function attested in all four languages to the morphosyntactic one found in Umpithamu. As already mentioned in Section 4.1, structures with inanimate Agents are special in that they take obligatory ergative marking for the Agent nominal (rather than optional marking, as for all other types of Agents). If an inanimate Agent affects a human Patient, moreover, there is a second set of features that sets apart such structures: The inanimate Agent is not cross-referenced with a nominative pronoun, and the human Patient is obligatorily cross-referenced, with an accusative or oblique pronoun.^[14] Some of these structures are illustrated in (25) and (26) below, for Umpithamu and Umbuygamu.

In other words, except for the obligatory use of an ergative-marked nominal, these structures are very similar to the impersonal structures discussed in the rest of this section.^[15] They are also interesting semantically, because they can be related to the lexical functions found in all four languages, as well as to the productive passive-like function found in Umpithamu, which would make them a potential bridge between the two. What links these structures to the idioms for involuntary physical processes is the typically involuntary involvement of the human Patient. Inanimates are either natural forces with unpredictable ‘generalized’ agency, or non-motive entities that are not typically monitored for agency (see Fauconnier 2012: 37–38 on some relevant grammatical differences). In either case, the activities they instigate more typically ‘overcome’ people than actions instigated by other humans, which are monitored for agency and are not generalized sources of undirected or unpredictable ‘spontaneous’ agency. In this sense, such activities are not unlike involuntary physical processes, which affect a human being beyond their control. On the other hand, there are also a number of links to voice mechanisms. Most generally, spontaneous events are a common historical source for passive constructions, as shown, for instance, in Haspelmath (1990). More specifically, what links structures like (25) and (26) to voice mechanisms is the fact that they have a very particular configuration of participants, which typically triggers some sort of voice or valency change in languages that have the relevant structures. From a broad typological perspective, inanimate Agents acting on human Patients run counter to the classic participant hierarchies often assumed to govern case marking (Silverstein 1976), and they are a configuration that tends to be associated with special voice options and related phenomena (like hierarchical alignment, compare Gildea and Zuñiga 2016). And in the immediate region, structures with inanimate Agents are often associated with the same valency-reducing morphology that characterizes passives, antipassives and the like, though sometimes with different case frames. This is illustrated in the Guugu Yimidhirr structure in (27) below, where the verb is marked for reduced valency with the reflexive suffix, in combination with standard transitive case frames for the arguments (vs. distinct case frames for canonical reflexives, see further in Verstraete 2011b).

In other words, the very minor category of inanimate Agents acting on human Patients could, by virtue of its formal and semantic links with the other categories, have served as a kernel or a ‘bridge’ towards the passive-like function in Umpithamu, as shown in Figure 2. This scenario remains speculative of course, but even if the origins of the passive-like use lie elsewhere, it is clear from the contact-induced origins of the structure that, unlike the lexical uses, the passive-like use was not borrowed along with the formal structures from Lamalamic, and must have originated later than these lexical uses.

Figure 2:

Scenario for development of middle voice system in Umpithamu.