Bribri media tantum verbs and the rise of labile syntax

3.1 The -r suffix as a passive voice marker

In Bribri, passive voice is expressed by the -r suffix, which demotes the A argument when applied to transitive verbs (e.g., chö́k ‘to say’, ppö́k ‘to hit’) and the S argument when applied to agentive intransitive verbs (e.g., tsö́k ‘to singʼ, ĩnṹk ‘to playʼ). Non-agentive intransitive verbs such as ‘to sleep’ or ‘to die’ cannot take this suffix (cf. Dickeman-Datz 1984: 118).

Passive voice is derived by adding the -r suffix to the same base as the remote perfective form of active voice verbs but with the loss of any final glottal stop (Constenla Umaña et al. 1998: 26; Margery Peña 2005: lxiv); see Table 1. To form the passive perfective, a low toned -ẽ suffix is added after -r.^[12] The passive imperfective has no special additional ending, and -r surfaces simply as [ɾ]. The infinitive ending of a passive voice stem is always -ũk, which consists of the low toned thematic vowel -ũ plus the infinitive ending -k. Table 1 also includes some examples of strong stem suppletion between infinitival and perfective forms (cf. chö́k ‘to say’ and ppö́k ‘to hit’).^[13] In Table 1, the roots for ‘sing’ and ‘play’ are agentive intransitive, while ‘say’ and ‘hit’ are transitive.

The suffix -r is used in three types of passive constructions: (i) Agentless Passive, (ii) Impersonal Passive, and (iii) Potential Passive. In (11) the transitive verb yö́k ‘to drink’ occurs in an active voice construction. The use of the -r suffix to suppress the most Agent-like argument of a transitive verb stem is shown in (12).

Example (12) is an example of the Agentless Passive Construction. The structural modifications of this construction compared to (11) are the obligatory omission of an overtly expressed Agent phrase, and the presence of the suffix -r on the verb. However, (12) does not appear to be an instance of a true promotional passive, i.e., balo’ in (12) does not become a true subject. If the absolutive argument balo’ in (12) were a true subject, we would expect it to be able to undergo subject-to-object raising in the same way as can S arguments of intransitive verb stems and A arguments of transitive verb stems. In Bribri, active voice transitive matrix verbs (i.e., tsö́k ‘to hearʼ in (13) and (14) and sãũk ‘to see’ in (15)) allow the raising of S/A arguments from complement clauses (CC) into the object of the matrix clause (see Dickeman-Datz 1984 and Section 6 for a detailed discussion).

In (13), the complement clause ‘Juan sings’ is indexed in the matrix clause by the 3sg dummy absolutive verbal agreement prefix i-. In (14), the S argument (i.e., ‘Juan’) of the intransitive verb stem tsö́k ‘sing’ in the complement clause is raised to object position in the matrix clause and leaves a zero in the complement clause from which it was raised. The intransitive verb stem of the complement clause must now appear in its infinitive form. By contrast, the single argument of a passivized verb form does not show the same subject properties of the single argument of an intransitive verb stem such as Juan in (14), judging by the ungrammaticality of (15).

The second type of passive construction involving the -r suffix is the Impersonal Passive. Compare (16) and (17).

We consider (17) to be what is often called impersonal passive (see Kulikov 2011: 376–379; Mal’chukov and Siewierska 2011: 37; Siewierska 1984: 93, 1988: 269; among others), in which the argument of an agentive intransitive verb is suppressed, and the resulting construction contains only the verb with the -r suffix, plus possible modifiers.^[15]

Another function covered by the -r suffix in Bribri is the expression of potentiality. In the Potential Passive construction, the verb appears with the -r suffix, and the argument that is able to perform the action described by the verb is encoded as an oblique, by the postposition ã ‘in, inside of, for, to, from’, as in (18).

Crucial to the passive analysis of the -r suffix is the fact that when this suffix is added to transitive verbs (such as chö́k ‘say’, tsö́k ‘sing’, ppö́k ‘hit’ kímũk ‘help’, tö́k ‘cut’, see Jara Murillo and García Segura 2009: 68–71 for a list), it does not convey the meaning of an event that occurs spontaneously, i.e., without the instigation of an agent. In (19), the transitive verb stem tsakö́k ‘to break’ appears in the active voice form with an ergative and an absolutive argument.

When tsakö́k takes the -r suffix, as in (20), no spontaneous occurrence can be assigned to the event. Even though the Agent cannot be overtly expressed in the construction, it is nevertheless implied that someone broke the pot.

According to our consultant, the only way to conceptualize the event of pot breaking as occurring spontaneously is as in (21), where the verb takes the -r suffix and the resultative-completive suffix -wã.^[16]

It is likely that the resultative meaning of the -wã suffix facilitates a spontaneous (anticausative) interpretation (the state of being broken is the result of a change of state conceived as spontaneous), as is often the case cross-linguistically (see Nedjalkov 1988).

The absence of a spontaneous interpretation for the -r suffix alone is observed also with other lexical verb stems. For instance, when the transitive verb stem diâũk ‘melt’^[17] in (22) combines with -r suffix, the event is not interpreted as occurring spontaneously, without the instigation of an external force. The parentheses around the directional suffix -ã́n ‘downwards, falling movement’ mean that this element can be omitted from the construction.

As happens with tsakö́k ‘break’, diâũk ‘melt’ must combine with the -r suffix and the resultative-completive suffix -wã in order to convey the interpretation of a spontaneous melting as in (23).

3.2 The -r suffix as an intransitive anticausative verbalizer

Another function of the -r suffix in Bribri is that of a verbalizer which derives intransitive anticausative verbs from adjectives and nouns. Examples (24) and (3), reproduced as (25), show that the adjectival root bâ ‘hot’ combines with the verbalizer -r, yielding ‘to become hot’. Example (26) shows that the nominal root nṹ ‘corpse’ combines with -r, yielding ‘to rot’.

The adjectival and nominal roots verbalized by -r always have their infinitive ending in -ũk. Table 2 offers some examples of this highly productive derivation (see Jara Murillo 1995: 18; Sánchez Avendaño 2009: 63 for additional examples). The first column in Table 2 indicates the syntactic category to which the root belongs. The second column illustrates the infinitive formation of derived intransitive anticausative verbs, where the verbalizer -r on a non-verbal stem is followed by the infinitive ending -ũk. Adjectival stems can undergo haplology during derivation (cf. chabáchabâ ‘muddy’ or sẽ̂ssẽ ‘cold’).

The peculiar status of the anticausative formation in Bribri requires a short diachronic excursus. In Bribri, causative verbs can be formed by creating a compound which features the verb ṹk ‘to make’, and, to its left side, one of the following elements: [N], [Adj], [N + N], or [N + Adj], as in bâũk ‘to heat sth.’, formed by bâ ‘hot’ plus ṹk ‘to make’ (see Table 2). All such verbs have an anticausative counterpart, where the -r suffix is added before -ũk as in bâ-n-ũk ‘to become hot’, yielding causative-anticausative pairs. However, synchronically, the causative stem ṹk ‘makeʼ has been reanalyzed as an infinitive ending and disappears when the anticausative verb is conjugated, leaving no trace of causative morphology. For this reason, forms such as bâ-n-ũk are analyzed in this article as cases where the -r functions as an intransitive anticausative verbalizer for adjectival and nominal roots (or a combination of these) and -ũk as the infinitive ending of the verbalized form.

In Table 2, all infinitives of derived intransitive verbs end in -n-ũk (morphophonologically /-ɾ-ũk/). As shown in (24)–(25), a verb such as bânũk ‘to become hot’, as all other verbs in Table 2, has two forms: bâ-r for the imperfective (no extra suffix) and bâ-n-ẽ for the remote perfective (-ẽ suffix). These two aspectual forms are pivotal in determining whether a verb is a derived anticausative. For instance, a verb like apáinũk ‘to shake from fear or cold’ might be taken, formally and semantically, as a derived anticausative verb if we only take into account its infinitive form. However, the [n] in the infinitive apáinũk must be part of the root, i.e., [apáin-ũk] (/apaíɾ-ũk/), and not a surface manifestation of the -r suffix. This can be seen by looking at the imperfective and remote perfective forms of this verb, which are [apain-ṹ] (with the imperfective intransitive suffix -ṹ of active voice verbs) and [apáin-ɛ̃Ɂ] (with the glottal stop found in the perfective of the active voice conjugation) respectively. If apáinũk were morphologically derived by means of the -r suffix, i.e., [apái-n-ũk] (/apái-ɾ-ũk/), then its imperfective and perfective forms would be *[apaí-ɾ] and *[apaí-n-ẽ] (without a glottal stop), respectively.

3.3 Morphemes coding reflexive and reciprocal functions

Bribri uses unbound pronominal forms to express reflexive and reciprocal functions: ẽ’ and (ã)ñĩ́ respectively.^[19] The reflexive and reciprocal are valence-decreasing syntactic constructions in Bribri. Usually, only syntactically transitive verb roots can occur in reflexive and reciprocal constructions: when they do, the transitivity value of their root must change to intransitive. To see this, consider the active voice verb stems yawö́k ‘to do’ and ppö́k ‘to hit with a long object’. Both of these take the transitive imperfective suffix -é in non-reflexive constructions, as shown in (27) and (28).

When these transitive verb stems appear in a reflexive (29) or reciprocal (30) construction they change their imperfective suffix from -é (transitive) to -ö́ (intransitive).

5.1 [non-abs]wã [abs] [v _{media tantum} ]

Two out of five mtl verbs, chẽ́-n-ũk ‘to be known, to knowʼ and mĩ́-n-ũk ‘to go, to take’, have their non-absolutive argument marked by means of the postposition wã. As explained in Section 2, this postposition encodes both possessors and ergative arguments in Bribri and as such represents a cross-linguistically common case of genitive/ergative polysemy known, for instance, from the pre-history of the Proto-Indo-European case system (Baerman et al. 2002; Schmalstieg 1981; Vaillant 1936; among others). The verb chẽ́-n-ũk means ‘to be known’ when used in a construction with a single core argument such as (5), reproduced as (31).

The verb chẽ́-n-ũk can also appear in two-argument constructions, as in (4), reproduced as (32).

One might wonder, given that the verb bears intransitivizing morphology in both the transitive and intransitive frames, whether the ergative postposition in (32) introduces an oblique phrase instead of a syntactic argument, i.e., whether the free translation of (32) is rather something like ‘The story is known of/by him’. This analysis is at least synchronically impossible for (32): Bribri does not allow the expression of an oblique Agent phrase with verbs taking the -r suffix with passive function, hence the ungrammaticality of (34) as opposed to (33). The substitution of wã with the other ergative marker tö to express the Agent phrase also leads to ungrammaticality, see (35).

Another labile verb which takes a non-absolutive argument marked by wã is mĩ́nũk.^[22] This verb means ‘to go’ when used in an intransitive construction with an unmarked absolutive as in (36).

The same verb means ‘to take (to/with)’ when employed in two-argument constructions such as (37).^[23]

As in the case of chẽ́nũk ‘to be known, to knowʼ in (32), one wonders if the free translation of (37) should be ‘He went to a place up there by his motherʼ. This analysis is again ruled out by the lack of passives with overtly expressed Agents in Bribri. Additionally, NPs marked by wã in constructions such as (37) are capable of controlling the reference of an anaphoric pronoun in a following postpositional phrase (see Section 6.3). If the phrase marked by wã in (37) were an oblique, it would not have this ability.

5.2 [non-abs]_{LOC PP} [abs] [v _{media tantum} ]

The mtl verb kiánũk means ‘to be wanted/neededʼ when in a construction with a single core argument as in (38).

The only possible interpretation of (38) is that someone wants or needs coffee, even though the Experiencer of this desire cannot be overtly expressed in the construction. This verb, and possibly chẽ́nũk ‘to be known, to knowʼ, likely go back to a morphological passive derived stem, but the corresponding active voice verb no longer exists. When the verb kiánũk occurs in a two-argument construction, it means ‘to want, to need, to desire’. The non-absolutive argument in (39) is marked by a locative postposition which elsewhere means ‘upon, on’ and implies contact with a surface.

The verb ṍnũk means ‘to fall’ or ‘to scream’ depending on the context, when found in a construction with one syntactic argument, as in (40).^[24]

In (41), the non-absolutive argument of the verb ṍnũk is marked by the locative postposition ã preceded by ẽ́r ‘liver’, yielding the complex postpositional phrase ẽ́n ã which means literally ‘in (one’s) liver’. In this context, the meaning of the verb is ‘to understand’, as in (41).^[25]

The semantic shift of a verb of motion directed downwards to a verb of cognition is not uncommon, cf. Sanskrit ava-gam ‘understand’ formed by ava ‘down’ + gam ‘go’. When used in two-argument constructions with an animate absolutive argument, ṍnũk changes its meaning to ‘to remember’, as in (42).

5.3 [non-abs] wã/loc pp [abs] [v _{media tantum} ]

The verb dalë́nũk ‘to feel pain, to respect, to love, to exert oneself’ can take a non-absolutive argument marked by either the ergative postposition/possessor marker wã or a lexically determined locative postposition, with a concomitant change in the meaning of the construction.^[26] The derivational history of this verb form can conceivably be traced back to the noun dalë́r ‘pain’.^[27] The verb dalë́nũk means ‘to feel pain’, when used in a construction with one argument, as in (43).

A meaning propositionally identical to (43) can be expressed by constructions such as (44).

This verb can be used in two types of two-argument constructions. In the first, the verb appears with a non-absolutive argument marked by the complex postposition ẽ́n ã ‘in (one’s) liver’ and an animate absolutive argument and means ‘to love, to miss’, as in (45).

Second, when employed with a non-absolutive argument marked by wã and an animate absolutive, dalë́nũk means ‘to respect’. In (46), the speaker is being interviewed about the process of becoming awâ, the shaman of the community. He describes how people show great respect to an awâ and dwell on finding the best food and hammock for him to sit or sleep in. In the first line, the distal demonstrative e’ refers to the awâ, and the pronoun i- in the second and third lines is co-referential with e’.

In the second and third clauses, the verb dalë́nũk appears with a pre-verbal absolutive coded by i- and a non-absolutive argument marked by the ergative postposition wã. It is noteworthy that dalë́nũk can combine with the reciprocal and reflexive pronouns añĩ́ (47) and ẽ’ (48).

The fact that dalë́nũk can be used in reflexive and reciprocal constructions arguably provides evidence in favor of the syntactic transitivity of the form dalë́-r in (47)–(48). If the verb form dalë́-r ‘to feel pain’ were not transitive (at least syntactically), it should not be able to occur in reflexive and reciprocal constructions that are syntactically intransitivizing in Bribri (see discussion in Section 3.3). However, because dalë́nũk exists only in combination with -r morphology, there is no overt formal change in the suffixes that the verb takes when found in a reflexive or a reciprocal construction versus the transitive construction, i.e., there is no analogous change from the imperfective transitive suffix -é to the imperfective intransitive suffix -ö́, as shown in Section 3.3. Data from oral tradition texts show that kiánũk ‘to be wantedʼ and ṍnũk ‘to fall, to screamʼ can also occur in reciprocal and reflexive constructions (see Jara Murillo 1993).

In this section, we showed how non-absolutive arguments in [non-abs] [abs] [v _{media tantum} ] constructions are formally marked in main clauses. In the next section, we focus on the behavioral properties of mtl verbs and their non-absolutive arguments.

6.1 “Complete” and “incomplete” subject-to-object raising in Bribri

Dickeman-Datz (1984: 122) shows that in Bribri only transitive and “indefinite voice” (media tantum in our terms) matrix verbs (i.e., verbs with the -r suffix in a two-argument construction) allow subject-to-object raising.^[30] By contrast, non-derived intransitive verb stems in the matrix clause never allow raising in this language. In Bribri, an embedded complement clause in absolutive position is usually terminated by the anaphoric/distal demonstrative pronoun e’ as in (49).

If the matrix clause verb is transitive, as sã́ũk ‘to see’ in (50), the embedded complement clause can be extraposed and replaced by the ‘dummy’ absolutive i- in the matrix clause. Note that the complementizer tö introducing the extraposed complement clause is homophonous with the ergative marker tö.

In the analysis of Dickeman-Datz (1984), after extraposition takes place, the A argument of the complement clause, e.g., ‘Trini’ in (51) can be raised to become the object within the matrix clause.

As shown in (51), when the A argument of the complement clause ‘Trini’ is raised to the object position in the matrix clause, it leaves a zero in the complement clause and the verb ‘eat’ appears in its infinitive form. This type of raising has been called “complete” raising (Dickeman-Datz 1984: 123). Complete raising can only occur when the verb stem in the matrix clause is syntactically transitive. The fact that A arguments and S arguments (cf. (14)) in complement clauses can undergo complete subject-to-object raising points to the existence of a nominative pivot in Bribri. This is further corroborated by the fact that P arguments cannot undergo (“complete”) raising. In (52), the absolutive argument ali’ ‘manioc’ of the transitive verb stem ñũk ‘eat’ in the complement clause is unsuccessfully raised to object position in the matrix clause, and the sentence is ungrammatical.

Dickeman-Datz (1984: 123) describes another kind of raising in Bribri, which she calls “incomplete” raising. She argues that S arguments of stative and indefinite voice verbs (i.e., verbs with the -r suffix and presumably one single syntactic argument) in complement clauses can only undergo incomplete raising. Further, she argues that S arguments of non-stative underived intransitive verb stems and A arguments of transitive verb stems in a complement clause can undergo incomplete raising where a semantic distinction exists between completely and incompletely raised forms. This is illustrated with the transitive verb stem ttö́k ‘to kill’ in the complement clauses in (53)–(54).^[32]

complete raising

incomplete raising

In (53) the A argument síkua ‘white man’ of the transitive verb stem ttö́k in the complement clause undergoes complete raising and becomes the object (i.e., absolutive) of the transitive verb stem tsö́k ‘hear’ in the matrix clause. In (54), incomplete raising occurs. The A argument (subject) síkua ‘white man’ of the complement clause, raised to the object of the matrix clause, receives a pronominal “copy” in the complement clause (i.e., ieʼ), and the verb in the complement clause appears in its conjugated remote perfective form, not in the infinitive as in (53). In terms of meaning difference, in (53), where the raising of the A argument is complete, the speaker did hear first-hand the shooting of the tapir. But in (54), where the raising of the A argument (‘white man’) is incomplete, the speaker did not hear himself the event of tapir killing; rather, he might have heard the white man talking about it. This possibly points to the existence of grammatical evidentiality in Bribri, manifested through the opposition between complete (directive) versus incomplete (inferential/indirective) raising with verbs of perception. This topic goes beyond the scope of the present article. While further research is needed to substantiate this claim, a distinction between firsthand and non-firsthand evidentiality is reported in other Chibchan languages such as Ika (Bergqvist 2012: 172; Frank 1990: 77).

According to our data, P arguments in Bribri can also undergo incomplete raising, as shown in (55).

The data in (55) contradicts the claim of Dickeman-Datz (1984) that “incomplete” raising can only target the S/A category in Bribri.^[33]

To summarize, both complete and incomplete raising require underived transitive verb stems in the matrix clause in order to occur. Complete raising can be taken as strong syntactic evidence for the subjecthood of the raised argument, because this operation is only available to S/A arguments in complement clauses (cf. the ungrammaticality of (52)). On the other hand, incomplete raising cannot be taken as syntactic evidence for the subjecthood of the raised argument, because this type of raising is available to S, A and P arguments in complement clauses indistinctively.

6.2 Behavioral properties of [v _{media tantum} ] in [non-abs] [abs] [v _{media tantum} ] constructions

In this section, we test whether mtl verbs allow subject-to-object raising (whether complete or incomplete) when they are in the matrix clause. We do so in order to see whether mtl verbs behave like transitive verb stems that allow this type of raising. This behavioral property can be tested only with those verbs that allow an entire complement clause as their absolutive argument. These are kiánũk ‘to want, to be wantedʼ, chẽ́nũk ‘to know, to be knownʼ, and ṍnũk ‘to fall, to understandʼ.

The mtl verb kiánũk ‘to be wanted, to want’ allows the extraposition of an embedded complement clause and its replacement with the dummy absolutive i- (56). In this respect, kiánũk ‘to be wanted, to want’ behaves precisely like the transitive verb stem sã́ũk ‘to see’ in (50).

This verb also allows, just like the transitive verb stem sã́ũk ‘to see’, complete subject-to-object raising. In (57), the raised A argument (i.e., ‘Trini’) leaves a zero in the complement clause and the verb appears in its infinitive form.

The verb ṍnũk ‘to fall, to scream, to understand, to remember’ allows only incomplete subject-to-object raising. In (58), the complement clause in absolutive position is extraposed and replaced by the dummy absolutive i-.

In (59), the raising is incomplete because the A argument (i.e., ‘white man’) which has been raised to become the object of the matrix clause must be pronominalized with a co-referential pronoun in the complement clause.

Incomplete subject-to-object raising is also the only option when chẽ́nũk ‘to be known, to knowʼ is the matrix clause verb. In this case too, it is possible to extrapose the complement clause and replace it with the dummy absolutive i-, as in (60).

The A argument of the complement clause, i.e., ‘Trini’ in (60), can then undergo incomplete subject-to-object raising in (61).

Recall that one of the requirements for either complete or incomplete raising to occur in Bribri is that the matrix verb be transitive (Dickeman-Datz 1984: 122). Examples (57), (59), and (61) show that kiánũk ‘to be wanted, to wantʼ, ṍnũk ‘to fall, to understandʼ, and chẽ́nũk ‘to be known, to knowʼ in a [non-abs] [abs] [v _{media tantum} ] construction pattern like transitive verb stems in a matrix clause in that they allow complete and/or incomplete raising of A arguments of transitive verb stems found in a complement clause. This evidence corroborates preliminary data and similar claims made by Dickeman-Datz (1984, 1985. However, only kiánũk ‘to be wanted, to wantʼ behaves entirely like a transitive verb stem in a matrix clause in that it allows complete raising of an A argument found in a complement clause (cf. (57)). In all examples in this section, the complement clause contains the same transitive verb stem ñṹk ‘eat soft things’ which must occur with two syntactic arguments, just like katö́k ‘eat hard things’ (see (6)). If chẽ́nũk ‘to be known, to knowʼ and ṍnũk ‘to fall, to scream, to understandʼ behaved entirely as transitive verb stems, they would allow complete raising of the A argument of the complement clause, just like kiánũk ‘to be wanted, to wantʼ.

6.3 Behavioral properties of [non-abs] in [non-abs] [abs] [v _{media tantum} ] constructions

In this section, we investigate whether non-absolutive arguments of mtl verbs can undergo complete subject-to-object raising when they appear with an absolutive and a non-absolutive argument in a complement clause. We do so to determine whether they show the same behavior as ergative phrases. We also determine whether non-absolutive arguments of mtl verbs can control the co-reference of an anaphoric possessive pronoun in a following possessive NP or postpositional phrase. This property is shared by the S/A category in Bribri.

Recall from Section 6.1 that, according to Dickeman-Datz (1984: 124), if the verb in the embedded complement clause is stative or in the indefinite voice (i.e., the verb has the -r suffix and presumably only a single syntactic argument), only incomplete raising of the S argument of such a verb can apply. Non-absolutive arguments in a [non-abs] [abs] [v _{media tantum} ] construction within a complement clause behave like S arguments in that they can undergo only incomplete raising. The fact that non-absolutive arguments of mtl verbs can undergo incomplete raising does not tell anything about their grammatical relation/syntactic behavior. This is because incomplete raising does not target a specific subset of grammatical relations in Bribri, but rather appears to be available to S, A, and P arguments indiscriminately (see Section 6.1).

For reasons of space, in (62) and (64)–(67) we only show the raising of the non-absolutive argument of each mtl verb from the complement clause into the object position within a matrix clause containing the transitive verb stem sã́ũk ‘to see’. For the same reason, we limit ourselves to showing that non-absolutive arguments cannot undergo complete raising only with the verb mĩ́nũk ‘to go’ in (63). All other mtl verbs do not differ from mĩ́nũk ‘to go’ in this respect.

The second behavioral property to be tested is the ability of the non-absolutive argument of mtl verbs to control co-reference of an anaphoric possessive pronoun in a following possessive NP or postpositional phrase. Only the S/A category has this property in Bribri. This is shown in (68) with the ergative marked A argument ‘Alíʼ, which is the only argument that can control co-reference of the anaphoric possessive pronoun ieʼ in the postpositional phrase ieʼ kipö́ ã. Notice that gender is not expressed in Bribri pronominal forms, so that i- in the examples below can refer either to a male or female referent.

In (68), if the hammock belongs to the woman, the possessor in the postpositional phrase would be expressed in one of two ways: either (i) by repeating the noun alâköl ‘woman’ before the noun kipö́ ‘hammock’; or (ii) by means of the demonstrative pronoun e’ ‘that’. In (68), the possibility that the possessive pronoun ieʼ is co-referential with a third person referent other than ‘Alí’ present somewhere else in discourse is ruled out. Example (69) shows that another third person referent in a postpositional phrase, i.e., ‘Hernán’, is not a possible antecedent for the possessive pronoun ieʼ.

If the hammock belongs to Hernán, this proper noun must be repeated before the noun kipö́.

Non-absolutive arguments of Bribri mtl verbs marked by the ergative postposition/possessor marker wã show the same ability to control co-reference of the following anaphoric possessive pronoun (i.e., i- or ie’) in a postpositional phrase ((70) and (71)) or in a possessive NP (72).

Non-absolutive arguments of mtl verbs marked by means of a locative postposition display the same ability to control co-reference as ergative-marked A arguments (and S arguments) have, see (73) and (74). This suggests that non-absolutive arguments of mtl verbs display yet another subject property.

Table 4 summarizes the coding and behavioral properties of Bribri [non-abs] [abs] [v _{media tantum} ] constructions. In Table 4, the symbol “–ˮ means that the corresponding property cannot be tested for some verb forms.

The verb kiánũk ‘to be wanted’ behaves the most like transitive active voice verbs in that it allows complete subject-to-object raising when found in matrix clauses. Non-absolutive arguments of mtl verbs can control the reference of a third person possessive pronoun in a following possessive NP or postpositional phrase but unlike S/A arguments they cannot undergo complete raising when found in a complement clause.

7.1 Possible origins of causative-anticausative oppositions

Recall from Section 3.2 that Bribri causative verbs are formed by placing a noun, an adjective, two nouns or a noun plus an adjective before the verb ṹk ‘to make’ (see Table 2 for some examples). All causative verbs formed this way have an anticausative counterpart, where the -r suffix is added before ũk. However, synchronically, there is no trace of causative morphology in the conjugated forms of the anticausative counterparts. Given that the causative derivational stem ũk going back to the verb ṹk ‘makeʼ is identical (homophonous) with the infinitive suffix -ũk one may posit the following diachronic scenario for the emergence of Bribri -r anticausatives. Once the verb ũk ‘make’ began being reanalyzed as an infinitive ending, speakers started to add -r after what they considered to be the root, i.e., the stem without -ũk, by analogy with active voice verbs ending in -ũk and -ök that do not participate in causative-anticausative alternations.^[37] Subsequently, this -r suffix was extended to all transitive verbs, to form passives. In support of this diachronic scenario, there is literature indicating that cross-linguistically anticausative morphemes may commonly develop passive and other intransitive functions (Haspelmath 1987: 35), eventually becoming middle voice markers. Although frequently the anticausative-to-passive (and anticausative-to-middle) path is only part of a larger diachronic pathway starting with the reflexive function (Haspelmath 1990), cases of direct development from anticausative to passive/middle are not unknown.^[38]

7.2 The rise of labile verbs

The syntax of Bribri mtl verbs is of special interest for a diachronic typology of (P-)lability. In Bribri, we only find a small class of labile verbs (a subclass of media tantum verbs – see Table 3) within the mass of non-labile verbs. This small island of lability suggests a diachronic scenario for the rise of labile verbs which, to the best of our knowledge, has been hitherto unnoticed in the typological literature. The core of this mechanism can be described as follows. The morphologically marked intransitive -r verbs started to be used in active transitive constructions to make up for the lack of a corresponding verb form without the -r suffix which would normally be used in an active transitive construction. In this scenario, some mtl verbs expanded their syntactic uses by analogy with the active voice paradigm. This is illustrated in Table 5, where “–” means lack of a form/function pairing.

The two-argument (quasi-transitive) use of mtl verbs (which, originally, presumably were only used in one-argument constructions) emerges by analogy with the productive system found with transitive verbs, which can participate in a passive (syntactically intransitive) construction by adding the -r suffix. This two-way system very likely prompted the expansion of some media tantum labile -r forms into the transitive domain to fill in the gap in their defective syntactic paradigm. We call this mechanism media tantum transitive expansion, for lack of a better term. However, mtl two-argument constructions have not reached full transitivity (yet), as evidenced by the lack of some subject properties of non-absolutive arguments in two argument constructions (see Table 4). The development outlined in Table 5 is quite different from the case of Latin deponent verbs, a lexically specified set of verbs whose morphologically passive forms are used in active but not passive constructions. In this sense, their syntactic paradigm remains defective. Unlike Latin deponent verbs that left an empty slot in the passive paradigm, Bribri mtl verbs have developed a two-argument (quasi-transitive) use.