Underspecification and ambiguity of voice markers

Authors’ draft, to appear as: Giomi, Riccardo & Guglielmo Inglese. 2024. Underspecification and ambiguity of voice markers: synchrony and diachrony. In Vagueness, ambiguity, and all the rest. Linguistic and pragmatic approaches, Ilaria Fiorentini & Chiara Zanchi (eds.), 110-147. Amsterdam & Philadephia: John Benjamins. Underspecification and ambiguity of voice markers: synchrony and diachrony Abstract Voice markers have a notorious cross-linguistic tendency towards multifunctionality, in that a given marker can encode more than one voice operation at a time, such as reflexive and passive. In addition, diachronic typological research has also shown that patterns of multifunctionality of voice markers historically come about following paths that are not necessarily unidirectional. Taking stock of these premises, in this paper we propose a new typology of voice markers grounded on the notion of underspecification and ambiguity, and, by adopting the perspective of Functional Discourse Grammar, we argue that the lack of unidirectionality in the grammaticalization of voice markers follows from their status as interface operators. Keywords: voice markers, ambiguity vs. underspecification, grammaticalization, Functional Discourse Grammar, typology 1. Introduction Grammatical voice can be defined as “a grammatical category whose values correspond to particular diatheses […]” (Zúñiga & Kittilä 2019:4), the latter being defined as specific mappings of semantic roles (e.g., Agent or Patient) onto syntactic functions (e.g., subject or direct object; see Kulikov 2011). In other words, diathesis refers to a particular argument structure, or valency 1 pattern, that differs in some way from the ‘basic’ one of a given verb, defined as “active” voice (see Bahrt 2021:18-19; on the relation between valency, transitivity and diathesis see Creissels forthc.). Possible diatheses, or voice operations, include, e.g., the reflexive and the passive. Following this definition, a voice marker is an overt exponent of grammatical voice on the verb. Two striking facts about voice markers are, on the one hand, their renown cross-linguistic tendency towards high multifunctionality and, on the other, their unusual (and recently rediscovered) property of being diachronically prone to various types of bidirectional functional changes. As regards the former, it is well known that languages often employ a single construction to mark two or more voice distinctions. Leaving aside valency-increasing operations like causativization and applicativization, well-known multifunctionality clusters include, among others, reflexive, reciprocal, anticausative, passive and antipassive functions.1 Such patterns of multifunctionality (aka ‘voice syncretism’) have been described in quite some detail, both within and across languages (see e.g. Geniušienė 1987, Haspelmath 1987, 1990, Kemmer 1993, Kulikov 2010, Zúñiga & Kittilä 2019, Bahrt 2021, Inglese 2022a, Creissels forthc.). These typological and descriptive studies have offered various classifications of voice operations into more discrete categories, based on a number of formal and functional parameters. However, existing accounts tend to focus on a few broadly defined voice functions and do not usually offer a more reasoned typology of the mutual relations entertained by the various argument-structure oppositions involved in voice syncretism. The first goal of this paper is to argue that such a typology can be attained by taking into account the semantic distinction between underspecification and ambiguity. 1 As valency-increasing voices less frequently participate in multifunctionality patterns than those that express valency decrease (see Bahrt 2021:161), we leave the former out of this study. 2 As regards their diachronic development, voice markers represent a remarkable exception to the well-known fact that the processes whereby new grammatical meanings emerge tend to be overwhelmingly unidirectional (e.g. Haspelmath 2004), as discussed most extensively in Bahrt (2021). The second goal of this paper is to account for the multidirectionality in the development of voice markers, and to connect this with our synchronic typology of valency-changing functions. The paper is organized as follows. In Section 2 we first offer an overview of non-valencyincreasing operations and of the syncretism patterns that have so far being documented for these functions in the languages of the world. In Section 3 we propose our classification of voice oppositions in terms of semantic underspecification and ambiguity and then move on to discussing the role of contextual and lexical information in determining the specific interpretation of multifunctional voice markers. In Section 4 we shift to a diachronic perspective, presenting a brief overview of the attested bidirectional changes in the domain of voice marking. We discuss the potential bridging contexts in which some of these changes may conceivably take place, focusing in particular on the role played by underspecification and ambiguity in the extension to new valency-related functions. In doing so, our goal is to bring together the proposed synchronic taxonomy of voice oppositions with the uncommon diachronic behavior of voice markers and to propose an explanation of why bidirectional changes are attested with a significant frequency for these markers. Section 5 presents the conclusions of the paper. 2. Towards a typology of voice syncretism 3 2.1. Voice markers and their functions Various taxonomies of diatheses, or voice operations, have been proposed in the literature (e.g. Givón 2001, Kulikov 2010, Zúñiga & Kittilä 2019, Bahrt 2021, Creissels forthc). Before delving into the description of individual non-valency increasing voice distinctions, it is important to clarify that, in what follows, we use the labels A and P not to refer to the semantic roles Agent and Patient, but rather as umbrella terms that identify the first and the second argument, respectively, of a transitive clause, while S refers to the sole participant of a monovalent clause (see Creissels forthc. for details). Following Haspelmath (2011), in turn based on Comrie (1989), such labels may be understood as syntactically defined comparative concepts for the analysis of argument-marking systems, both cross-linguistically and intra-linguistically (across constructions). When we need to specify the semantic roles of A/P/S arguments, labels such as Agent, Patient, Experiencer etc. are used instead. In general, voice operations may be divided into three main types, based on their effect on valency with respect to the default transitive construction of the verb. 1. Argument-structure preserving (see Section 2.1.1): these operations alter the syntactic valency of the verb by demoting/promoting an argument to a more peripheral/central role, leaving the semantic valency unaltered; an example of this group is the passive voice, in which the P argument of the active voice is promoted to subject and the original A subject is demoted to oblique, e.g. the cat is chased by the dog. 2. Argument-structure modifying (see Section 2.1.2): these operations add or suppress a participant from the semantic valency of the verb (and hence an argument from the syntactic 4 valency), or alter the semantic role of one of the arguments (which is accordingly also assigned a more peripheral syntactic role); an example is the anticausative voice, in which the P argument of the active voice is promoted to subject and A is eliminated from the semantic valency of the verb, e.g. Italian il vaso si è rotto (*dalla bambina), ‘the vase broke (*by the girl)’. 3. Argument co-reference marking (see Section 2.1.3): these operations signal some sort of coreference between verbal arguments, thus assigning two semantic roles to the same participant; an example is the reflexive voice, in which the roles of A and P are typically attributed to a single NP encoded as subject, e.g. Mark sees himself in the mirror. In the reminder of this section, we illustrate a number of diatheses belonging to the groups 1 to 3. The goal is not so much to offer a full-fledged typology of voice (on which see especially Zúñiga & Kittilä 2019 and Creissels forthc.), but rather to provide explicit definitions and set up the stage for the discussion of underspecification and ambiguity in the voice domain (Section 3) and its diachronic implications (Section 4). 2.1.1. Argument-structure preserving diatheses To begin with, let us consider the types of diathesis shift in 1, which enact a reassignment of the grammatical relations of verbal arguments without affecting their semantic roles. This is the case of canonical passives and antipassives in (1) and (2). In (1), the active verb amant requires an A argument in the nominative case and a P argument in the accusative case; conversely, with passive amantur the P of the corresponding transitive clause is morphosyntactically realized as the subject of an intransitive clause, that is S, whereas the original A is demoted to an oblique phrase. Likewise, in (2a-b) the semantic roles of the two arguments are unaffected by 5 antipassivization, but their syntactic functions change: in (2a) A and P are flagged by the ergative and the absolutive case, respectively; by contrast in (2b) the original A takes absolutive case, thus functioning as S, whereas the original P is marked as an instrumental NP. (1) Passive: Latin (Indo-European) ita amant (…) quae ex se natos REL.NOM.N.PL from REFL born.ACC.M.PL thus et ab eis ita amantur and by 3PL.ABL thus love.PRS.PASS.3PL love.PRS.3PL ‘Which [i.e. certain animals] love their offspring, and are so loved by them.’ (Cic. Amic. 27) (2) Antipassive: Dyirbal (Pama-Nyungan: Dik 1983:242) a. b. bayi bargan baŋgul yaṛangu dyurganyu DET.ABS wallaby.ABS DET.ERG man.ERG spear.NFUT bayi yara baŋgul bargandu DET-ABS man-ABS DET-INSTR wallaby-INSTR dyurga-ŋa-nyu spear-ANTP-NFUT Both: ‘The man is spearing the wallaby.’ Related to the passive proper is the non-promotional passive, which features syntactic demotion of A, with P retaining its coding features as in the corresponding transitive construction: 6 (3) Non-promotional passive: Ukranian (Indo-European, Balto-Slavic: Sobin 1985:658) Cerkv-u zbudova-n-o bu-l-o ACC.SG COP-PST-N.SG church(F)- build-PTCP.PASS-IMPS.N.SG v 1640 roc’i (Lesev-ym) PREP 1640 year Lesiv-INS.M.SG ‘There was built a church in 1640 (by Lesiv).’ 2.1.2. Argument-structure modifying diatheses The diatheses in 2 alter the constellation of semantic and syntactic roles associated with transitive verbs. Consider first the anticausative function, illustrated in (4a-b). With respect to the active transitive verb fregit in (4a), which features a nominative A and an accusative P, encoding the Agent and the Patient, respectively, the corresponding anticausative verb frangitur in (4b) indicates a spontaneous event with the Patient coded as S. Semantically, this diathesis also implies that the Agent is removed from the event frame (and hence cannot be expressed): (4) Anticausative: Latin (Indo-European) a. Si quis (…) os fregit if INDEF.NOM bone(N).ACC.SG break.PST.3SG ‘If someone broke (someone else’s) bone.’ (Cato Orig. fr. 81, 1) b. frangitur aestus break.PRS.ANTC.3SG tide(M).NOM.SG ‘The rolling tide breaks’ (Lucr. 6,121) 7 Two other voice distinctions display a similar manipulation of grammatical relations with respect to the transitive construction (A suppressed, P changed to S): these are the potential passive or facilitative (e.g. French Ce livre s’est bien vendu (*par les livraires), ‘This book sold well (*by the booksellers)’; Guentchéva & Rivière 2007:570) and the so-called agentless passive (e.g. Portuguese Vendem-se casas, ‘Houses are sold’, Peres & Móia 1995:237). A construction that shares features of both agentless and non-promotional passives is what we label subjectless impersonal: in this diathesis the expression of an A argument is disallowed, but the P still retains the same properties as in the corresponding transitive clause and therefore does not command verbal agreement (e.g. Portuguese Vende-se casas).2 The crucial difference between anticausatives and facilitatives/agentless passives/subjectless impersonals is that the former do not entail the existence of an external Agent (whence their spontaneous semantics, cf. Haspelmath 1987), while the latter are always understood as involving a generic (typically, human) Agent. Nevertheless, we take the fact that these diatheses typically do not allow the syntactic realization of the A of the corresponding transitive construction as sufficient evidence to treat them as A-suppressing, on a par with anticausatives, contra their usual classification as sub-types of passives (cf. Zúñiga & Kittilä 2019:84−85,100). 2 Facilitative and subjectless impersonal diatheses may also apply to monovalent verbs (see Neshcheret & Witzlack- Makarevich 2016), but for reasons of space we do not discuss such cases here. Other voice distinctions that we do not address here include the self-benefactive and the indirect reflexive, which arguably form a single diathesis (Inglese 2022a:504) and have sometimes been analyzed as valency-increasing (Boeder 1968, Shluinsky 2021:415). 8 Specular to agentless passives are patientless antipassives, also known as “absolute” (Geniušienė 1987:83), “deobjective” (Haspelmath & Müller-Bardey 2004:1132) or “absolute antipassive” (Bahrt 2021:25). Unlike canonical antipassives, these constructions do not license the expression of a P argument, as shown in (5):3 (5) Patientless antipassive: Udmurt (Uralic, Permian: Geniušienė 1987:315) a. Puny vanz-es kurtčyl-e. dog.NOM all-ACC bite-3.SG ‘The dog bites everybody.’ b. Puny kurtčyl-iśk-e. dog.NOM bite-DEOBJ-3.SG ‘The dog bites.’ Two diatheses that we likewise ascribe to group (ii) are the deagentive, also referred to as converse, and the depatientive. These diatheses have in common that, besides a reassignment of grammatical relations, they also entail a distinct construal of the participants involved, typically as regards their degrees of control and/or affectedness. In this respect, these diatheses also entail a change in the semantic transitivity (understood as per Hopper & Thompson 1980) of the corresponding unmarked active voice construction. In the deagentive, or converse, construction 3 Note that, syntactically, antipassivization works differently based on the syntactic alignment of the construction to which it applies: in ergative alignment, antipassivization assigns the grammatical relation S to the A of the transitive clause, cf. (2b), whereas in nominative alignment this participant retains its subject marking, cf. (5b). 9 in (6b), the P featured in (6a) is coded as S, whereas the original experiential A is now construed as a Beneficiary.4 (6) Deagentive: Russian (Indo-European, Balto-Slavic: Babby 1999:45) a. My vspomnil-i staruju we.NOM remembered-PL old.ACC.F.SG pesnju song(F).ACC.SG ‘We all remembered the/an old song.’ b. Nam vspomnil-a-s’ staraja we.DAT remembered-F.SG-DEAG old.NOM.F.SG pesnja song.(F).NOM.SG ‘The/an old song came to our mind.’ (Lit. “the/an old song remembered-s’ to us”) The depatientive diathesis is similar to the antipassive (and in fact is regarded by some authors as a sub-type thereof) in that the A retains its morphosyntactic status, but differs in that not only is the P syntactically demoted, but its semantic role is also altered. For example, in (7b) the 4 As pointed out by Zúñiga & Kittilä (2019:88), the deagentive/converse diatheses are mostly restricted to experiencer verbs, that is, to predicates that do not take prototypical Agents and Patients as their arguments but rather Experiencers and Stimuli. Also note that the semantic and syntactic role assigned to the non-subject argument is remarkably variable: while in (6b) there is a Beneficiary encoded as a dative NP, in Sobaka ispugala-s’ groma (‘The dog(NOM) got afraid of the thunder(GEN)’; Kulikov 2010:380) the Stimulus of the active clause takes genitive case. At any rate, insofar as the oblique argument is less actively involved in the situation than the A of the corresponding transitive clause, there is reason to keep the deagentive/converse diathesis separate from the canonical passive. 10 original P kayŋ ‘bear’ is construed as a Goal/Direction (an alteration which is often captured in terms of low affectedness of the Patient in the literature, e.g. Vigus 2018): (7) Depatientive: Chukchi (Chukotko-Kamchatkan; Janic & Witzlack-Makarevich 2021:17) a. ətləg-e keyŋ-ən penrə-nen. father-ERG bear-ABS attack-AOR.3SG>3SG ‘Father attacked the bear.’ b. ətləg-ən penrə-tko-gʔe kayŋ-etə. father-ABS attack-DEPAT-AOR.3SG bear-DAT ‘Father rushed at the bear.’ 2.1.3. Argument co-reference marking diatheses The diatheses in group 3 are the prototypical reflexive and reciprocal. With respect to the transitive construction in (8a), the reciprocal and reflexive diatheses in (8b-c), signaled by the suffix -ž-, assign both Agent and Patient role to the same participants, jointly realized as a plural intransitive subject (whence the term duplex diathesis in Zúñiga & Kittilä 2019:153): (8) Reflexive and reciprocal: Tuvan (Turkic; Kuular 2007:1165,1213) a. Ava-m ača-zə-n mother-my father-her-ACC kuspakta-p tur embrace-CONV PRS.3SG ‘My mother is embracing her father.’ b. Ava-m bile mother-my and 11 ača-zə kuspakta-ž-əp tur-lar father-her embrace-REC-CONV PRS-3PL ‘My mother and her father are embracing each other.’ c. Boraxirilee-ler dovurak-ka bora-ž-er sparrow-PL dust-DAT dirty-REFL-NPAST ‘The sparrows dirty themselves in dust.’ While the function of reflexives and reciprocals is relatively straightforward, more problematic is the classification of so-called autocausative verbs (Creissels 2006:10, also known as endoreflexives, Haspelmath 1987:27). With this label, we refer to an interpretation of voice markers peculiar to non-translational motion events e.g. ‘rise’ or ‘turn’, as illustrated in (9): (9) Autocausative: Hittite (Indo-European; Inglese 2020:511,509) a. n=ašta GIŠḫulugannin EGIR-pa neyanzi CONN=PTC cart.ACC back turn.PRS.3PL ‘And they turn the cart back.’ b. LÚ.MEŠḪÚB.BÍ nēanda dancer(PL) turn.PRS.ANTC.3PL ‘The dancers turn around.’ In (9a), the form neyanzi is a transitive active verb that takes an Agent and a Patient argument flagged as A and P, respectively; the corresponding autocausative verb nēanda in (9b) takes only one S argument with the same coding as A in the transitive construction. The individuation of the diathesis shift underlying autocausative constructions crucially depends on the interpretation of the semantic role of S. Clearly, with respect to (9a), the autocausative has the effect of 12 removing an external Agent, thus triggering an effect similar to anticausativization. However, the remaining argument is not a Patient: in (9b), ‘the dancers’ rather displays properties of an Agent. In this respect, one could also analyze this construction as featuring suppression of the Patient, thus like a patientless antipassive. Both analyses fail to notice that the sole argument of the autocausative in (9b) shows properties of both Agents and Patients, as is typical of participants of verbs of motion (namely, it is a volitional participant initiating an event, but it also undergoes a change in location/position). Therefore, one could also analyze it as a type of reflexive. However, there are reasons to believe that autocausatives are semantically and syntactically distinct from reflexives. To illustrate this difference, let us consider the behavior of non-translational motion verbs in Italian. In this language, most transitive verbs license either reflexive voice marking via the clitic pronoun si or a non-valency-changing reflexive construction with the pronoun se stesso ‘oneself’ (e.g. colpirsi/colpire se stesso ‘hit oneself’). Non-translational motion verbs, however, typically (if not obligatorily) trigger an autocasative interpretation when occurring in the intransitive si construction (e.g. girarsi, ‘turn around’), but are not easily compatible with a ‘proper’ reflexive reading. This explains why they hardly ever occur in explicitly reflexive transitive constructions (cf. *girare se stesso, lit. “turn oneself”). For these reasons, we believe that autocausatives cannot easily be subsumed under a single diathesis but may need to be split over at least two separate diatheses. The diachronic implications of this particular status of autocausatives will be discussed in Section 4. 13 2.1.4. A typology of diatheses: summary Summing up, at least twelve non-valency-increasing functions of voice markers can be distinguished, each of which features a unique mapping of semantic roles onto grammatical relations. As we have already highlighted, some of these diatheses show strong affinities, to the point that they can be understood as instantiating a single, more abstract type of diathesis shift – that is, as expressing one and the same underlying grammatical function. In fact, this reasoning can be taken one step further, and the diatheses described in 2.1.1 to 2.1.3 can be lumped together into fewer, more abstract types; the constraint to this lumping is precisely that diatheses which cannot be reduced to a similar type of mapping between semantic and syntactic roles should never be assigned to the same group. Keeping this principle in mind, we arrive at the following, more fine-grained taxonomy of six clusters of diathesis shifts, which partly cut across the three larger groups presented in 2.1.1-2.1.3. Note that, in the remainder of this section, we understand the labels Agent and Patient in the broader sense of semantic macro-roles (Van Valin 2005), i.e. generalized semantic roles that cross-linguistically tend to be encoded as A and P arguments, respectively (Bickel 2011), thus encompassing more specific semantic roles than prototypical Agents and Patients sensu stricto. First, the anticausative (4b), facilitative and agentless passive can be brought together with the deagentive/converse (6b) in a single group of Agent-suppressing or -demoting diathesis shifts: as shown in (10), this diathesis cluster is characterized by the assignment of the grammatical relation S to a Patient(-like) argument and the absence of an Agent in underlying semantic structure. This explains why overt reference to an Agent is impossible in these diatheses, not only when the event is presented as a spontaneous change of state (anticausative) 14 or the A of the corresponding transitive clause is conceptualized as a non-agentively involved participant and expressed as an oblique argument (deagentive), but even when an active initiator would be logically presupposed by the event-type in question (facilitative and agentless passive, see 2.1.2). The same reasoning applies to the group of Patient-suppressing or -demoting shifts illustrated in (11), which subsumes the patientless antipassive (5b) and the depatientive (7b). Contrary to the previous cluster, these are characterized by the presence of an Agent(-like) argument (expressed morphosyntactically as S) and the absence of a Patient in underlying semantic structure: accordingly, it is the expression of a second participant (if present) as a P argument that is disallowed in this group of diatheses. Again, this is not only true of participants that somehow undergo the action denoted by the predicate but are not conceptualized as prototypical Patients (and are therefore expressed as oblique arguments, as in the depatientive) but also of participants which are logically implied to function as prototypical Patients (like the people suffering the dog’s biting in the patientless-antipassive example (5b)). Next, we have the cluster of non-promotional diathesis shifts in (12), subsuming the nonpromotional passive (3b) and the subjectless impersonal (see 2.1.2). This group differs from the others in that the Patient of the corresponding transitive construction retains its grammatical marking as P, but the Agent is either absent from underlying semantic structure and therefore cannot be overtly expressed even if logically implied (subjectless impersonal), or can only be expressed as an oblique argument (as in the case of the non-promotional passive, which in this respect is therefore similar to a canonical passive). The fourth cluster that subsumes more than one type of voice opposition is that of duplex diathesis shifts (13), comprised of the reflexive (8c) and the reciprocal (8b). As explained above, 15 these two diatheses are similar in that, semantically, they entail the assignment of both the Agent and the Patient role to the same participant(s), but morphosyntactically they feature as intransitive constructions with the single argument flagged as S. In this typology, only the passive and the antipassive diatheses cannot be assigned to any larger cluster, because they are both characterized by a unique mapping between semantic roles and grammatical relations. Thus, they each form a separate subset of diathesis shifts of their own. (10) Agent-suppressing/-demoting diathesis shifts (anticausative, facilitative, agentless passive, deagentive/converse) (11) (12) 5 Semantic roles (≠Agent)5 Patient Syntactic functions (Obl) S Patient-suppressing/-demoting diathesis shifts (patientless antipassive, depatientive) Semantic roles Agent (≠Patient) Syntactic functions S (Obl) Non-promotional diathesis shifts (subjectless impersonal, non-promotional passives) Here and below, the brackets around syntactic functions indicate that the corresponding semantic argument is not obligatorily expressed but, if it is, it always bears the syntactic function indicated within brackets. The brackets around semantic functions, however, are only used in those diathesis clusters which conflate more than one individual diathesis, and have a slightly different meaning: they indicate that the bracketed semantic argument is only present in some of the individual diatheses included in the cluster. The symbol ≠ indicates “semantic (macro)role different from”. 16 (13) Semantic roles (Agent) Patient Syntactic functions (Obl) P Duplex diathesis shifts (reflexive, reciprocal) Semantic roles Agent1 Syntactic functions (14) (15) Patient1 S Passive Semantic roles Agent Patient Syntactic functions (Obl) S Semantic roles Agent Patient Syntactic functions S (Obl) Antipassive As we discuss in Section 3.1, the relationship between different diathesis clusters, as well as the relationship among the various diathesis within each cluster, can be profitably understood in terms of underspecification and ambiguity. Before we turn to this issue, however, we need to take a closer look at the patterns of non-valency-increasing voice syncretism attested in the world’s language. 17 2.2. Voice syncretism patterns Earlier typological studies have repeatedly pointed out that voice markers often express more than one diathesis, which Bahrt (2021) refers to as voice syncretism, and that non-valencyincreasing functions tend to be co-expressed more frequently than valency-increasing ones (e.g. Nedjalkov & Silnitsky 1973, Geniušienė 1987, Haspelmath 1987, Oikonomou & Alexiadou 2022). For example, the Mediopassive inflection in Latin had at least passive (1) and anticausative (4) uses, and the Tuvan suffix -Iš- (surfacing as -ž-) may express both reflexivity and reciprocity, as in (8). Although such syncretism patterns have been discussed many a time, it is only recently that typologists have taken interest in assessing which patterns of syncretism are actually realized in the world languages (and why these patterns exist and not others). A detailed discussion of possible syncretism patterns is found in Bahrt (2021:Ch. 6). Generalizing over a sample of 222 languages, Bahrt finds that voice markers that express reflexivity are the most likely to express additional functions, followed by anticausative markers, with passive, reciprocal and antipassive markers being less prone to multifunctionality (see Figure 1).6 Figure 1: Tendency towards voice syncretism (Bahrt 2021:161) 6 Note that Figure 1 also includes the valency-increasing diathesis causative and applicative, which are not taken into account here (see fn. 1). 18 Turning to individual patterns, the data in Bahrt’s sample (2021:162) shows that reflexivereciprocal syncretism is the most frequent, followed by reflexive-anticausative, reciprocalanticausative, passive-anticausative, and passive-reflexive. Similar results are independently achieved by Inglese (2022a), who analyzes syncretism in a sample of 129 languages, and finds that reflexivization and anticausativization indeed constitute the most frequent syncretic functions. Inglese systematizes his findings by means of a semantic map, reported in Figure 2. As the map shows, the anticausative diathesis in particular seems to play a crucial role in syncretism patterns (it is the only one to be directly connected with every other diathesis in the map). By virtue of the connectivity hypothesis (Croft 2003:133), the map also predicts which syncretic patterns are expected to be cross-linguistically unattested (e.g., no marker should feature antipassive-facilitative syncretism unless it also expresses the anticausative diathesis). Figure 2: A semantic map of voice syncretism (from Inglese 2022a:508) 19 Overall, both Bahrt (2021) and Inglese (2022a) offer empirical support to the idea that voice syncretism is cross-linguistically pervasive, but also show that some syncretism patterns occur more frequently than others, and yet others are not attested at all (for some explanations of why this is the case see Inglese 2022b). It must be remarked that typological studies of voice syncretism, with the notable exception of Geniušienė (1987), tend to operate with a simplified typology of diatheses (only 5 in Bahrt 2021 and 8 in Inglese 2022a), so that we still lack systematic cross-linguistic data on syncretism patterns involving some of the diatheses introduced in Section 2.1. This is due on the one hand to a more coarse-grained classification of diatheses as typologically viable comparative concepts (e.g. the antipassive, patientless antipassives and depatientive defined in Section 2.1 are treated as one single antipassive diathesis by Bahrt 2021 and Inglese 2022a) and on the other hand to the lack of reliable data in descriptive grammars on individual diatheses. Nevertheless, some cross-linguistic tendencies can be detected in the literature also for some of the “minor” diatheses discussed in Section 2.1: for example, it appears that non-promotional diatheses may only be expressed by markers that also encode canonical passives (Geniušienė 1987:279). 3. Underspecification and ambiguity in voice syncretism 3.1. Underspecification and ambiguity In the literature on voice syncretism, scholars have variously addressed the commonalities and differences among individual functions, for instance by discussing how anticausatives differ from passives etc. (e.g. Haspelmath & Müller-Bardey 2004, Geniušienė 1987). However, the 20 multifunctionality of voice markers is usually regarded as a unitary grammatical phenomenon, without any attempt to distinguish between different types of multifunctionality. An explicit characterization of the relations among diatheses in terms of underspecification and ambiguity has, to our knowledge, only been proposed for reflexives and reciprocals (see Palmieri 2020 and references therein), so that we still lack a general understanding of how this distinction applies to the voice domain as a whole. In Section 2.1, we argued that the various functions of voice markers can be grouped into more general clusters, each one related to a different type of diathesis shift. We would now like to suggest that a more insightful understanding of the mutual relations among individual functions, both within and across these groups, can be achieved by distinguishing between different types of multifunctionality. The crucial notion of our proposal is the one that forms the main topic of this volume: the opposition between underspecification and ambiguity, that is, from an onomasiological perspective, between monosemous (underspecified) and polysemous (ambiguous) voice markers. As per Wasow (2015:32), we define ambiguity as follows: “Ambiguous expressions have more than one distinct meaning; […]. If expressions are thought of as picking out regions in some semantic space, then ambiguous expressions pick out more than one region.” Conceptual underspecification, or generality, can be defined, following Magni (this volume) as “the imprecision associated with any lexical item that selects a region of the semantic space with sharp but broad boundaries, has a general denotation, and lacks specificity regarding various dimensions of its semantic structure”. While Magni’s definition mostly concerns lexical items, we propose here to extend it to grammatical items as well. 21 Now, if one conceives of voice as a macro-functional space, the six diathesis clusters in (10)-(15) can accordingly be understood as more discrete “regions” of this functional space. This entails that voice markers that can express functions belonging to more than one diathesis-cluster are ambiguous (i.e. truly polysemous) with respect to distinct grammatical functions. For instance, this will be the case of a marker that has both Agent-suppressing and Patientsuppressing uses. Correspondingly, voice markers that can receive more than one interpretation belonging to a single cluster, say anticausative and facilitative or reflexive and reciprocal, will be underspecified (monosemous) between those specific interpretations. This is because, while each diathesis cluster can be clearly distinguished from the others in terms of semantic valency and/or syntactic role(s) of the relevant argument(s), the alignment between semantic and syntactic roles is constant across the specific functions that realize the same type of diathesis, as shown in (10)-(15). Hence, each of these interpretations should be thought of as a different “sense” of a single grammatical function. To illustrate how this model might work, we now consider some of the tests that have been proposed in the literature to distinguish between ambiguous and underspecified (or vague)7 expressions. Although these tests are usually applied to lexical elements, we believe that at least two of them are also valid for grammatical forms. These have been variously referred to in the literature. The first test is termed contradiction in Zwicky & Sadock (1975) and Kennedy (2019) and contrast in Murphy (2010); the second one goes by the name of zeugma (Murphy 2010) or identity of sense (Lakoff 1970, Zwicky & Sadock 1975, Kennedy 2019). As both these tests rely upon the (un)acceptability of the use of the construction under analysis in specific contexts, given 7 Note that not all of the authors referred to below distinguish explicitly between underspecification and vagueness (see also Magni, this volume). 22 the limited scope of this paper we are not in a position to test their validity in a systematic way, either for individual languages or on a large cross-linguistic scale. Here, we limit ourselves to illustrating the workings of these tests by means of data on the syncretic voice marker si in Italian.8 3.1.1. Underspecification and ambiguity: the contradiction test The contradiction test consists in “determining whether the same string of words […] can be used to simultaneously affirm and deny a particular state of affairs” (Kennedy 2019:242). If this is the case, the expression in question is truly polysemous, i.e. it has two clearly distinct meanings – because, if it were two senses of a single underlying meaning that were at stake, a contradiction would arise. As an example, consider (16): (16) Sterling’s cousin used to make people laugh with everything she did, though she was never in any way strange or unusual. She was funny without being funny. (Kennedy 2019:238) The sentence in boldface is acceptable because the term funny, which occurs twice, is used once with the meaning ‘hilarious’ and once with the meaning ‘strange’. This does not generate a contradiction, precisely because these two readings are not two senses of one and the same basic notion but pinpoint two genuinely distinct regions of semantic space. The other way round, the 8 For the sake of simplicity, we do not gloss the Italian examples in Sections 3.1, 3.2 and 4.2. Since the arguments made in these sections concern the interpretation of the examples, rather than their morphosyntactic shape, we believe that the translations proposed are sufficient to illustrate the points under discussion. 23 word cousin is underspecified between the readings ‘male cousin’ and ‘female cousin’: therefore, a sentence like Since Lily is a girl, she is Sterling’s cousin but not his cousin is utterly contradictory. In light of this distinction, we predict that two uses of a voice marker that instantiate different types of diathesis shift can co-occur in the same sentence, with reversed polarity, without resulting in a contradiction. Conversely, this is not possible for diatheses that belong to the same diathesis cluster, since these stand in a mutual relation of underspecification, not of ambiguity. The two sets of sentences in (17)-(18) lend support to this prediction: (17) a. Agentless passive (Agent-suppressing/demoting) vs. reflexive (duplex) Gli italiani non si governano, si governano da soli. ‘You don’t govern Italians, they govern themselves.’ b. Subjectless impersonal (non-promotional) vs. reciprocal (duplex) Da ogni parte si critica i regimi totalitari; soltanto loro non si criticano a vicenda. ‘Totalitarian regimes are criticized from every side; it is only them that do not criticize each other.’ (18) a. Anticausative/agentless passive vs. facilitative (both Agent-suppressing/demoting) ??Il condizionatore non si spegne, si spegne facilmente. Intended: ‘The air conditioner does not turn off by itself [anticausative], it is easily turned off.’; Or ‘‘One doesn’t turn off the air conditioner [agentless passive], it is easily turned off.’ b. Reciprocal vs. reflexive (both duplex): 24 ??I due pugili non si sono colpiti, si sono colpiti da soli. Intended: ‘The two boxers did not hit one another, they hit themselves.’ 3.1.2. Underspecification and ambiguity: the zeugma test While the contradiction test yields positive results for ambiguity and negative ones for underspecification, the zeugma or “identity of sense” test does exactly the opposite. This test consists in constructing a sentence in which two different interpretations are simultaneously activated for one and the same occurrence of a multifunctional expression. Such sentences can be constructed either by introducing an adjunct that activates a second sense of the multifunctional construction, or by coordinating two predicates or arguments each of which only makes sense on one of the two target-interpretations. If no contradiction arises, these two interpretations stand in a relation of monosemy/underspecification, otherwise they instantiate two separate meanings, in a relation of polysemy/ambiguity. Once again, the prediction is borne out: the sentences in (19)-(20) are acceptable because the two functions of the si-construction belong to the same cluster of diatheses, while those in (21)-(22) are ruled out because they simultaneously activate two interpretations pertaining to different clusters. Notably, at least for the reflexive-reciprocal syncretism, the existence of contexts that allow ‘mixed’ readings, that is, readings in which both the reflexive and the reciprocal meaning may be simultaneously activated (even though they may apply to different subsets of the entities involved), has been noted for a number of languages (see e.g. Murray 25 2008).9 This phenomenon, however, had not been previously observed for other diathesis clusters, to the best of our knowledge. (19) Adjunct (same diathesis) a. Agentless passive vs. facilitative (both Agent-suppressing/demoting) Si vendono libri, e anche bene! ‘Books are sold, and well too!’ b. Reflexive vs. reciprocal (both duplex) I due pugili si sono colpiti sia da soli che l’un l’altro. ‘The two boxers hit both themselves and each other.’ (20) Coordination (same diathesis) a. Reciprocal vs. reflexive (both duplex) Le due fazioni si sono confrontate tra loro e dichiarate indipendenti. ‘The two factions discussed with each other and declared their independence.’ b. Agentless passive vs. anticausative (both Agent-suppressing/demoting) Il forno si accende manualmente e spegne automaticamente. ‘The oven is turned on manually and turns off by itself.’ 9 Concerning Italian si, Palmieri (2020) argues that mixed readings are predominantly available with inherently reflexive verbs such as ‘wash’, but her experimental data also show that mixed readings are also, albeit more rarely, accepted for non-inherently reflexive verbs, e.g. ‘punish’. This points to the role of lexical information in disambiguating the reading of syncretic markers, as we discuss in Section 3.2. 26 (21) Adjunct (different diatheses) a. Anticausative (Agent-suppressing/demoting) vs. reciprocal (duplex) ??Si uccisero cadendo in un burrone, anche l’un l’altro. Intended: ‘They got killed falling in a ravine, and also killed one another.’ b. Agentless passive (Agent-suppressing/demoting) vs. depatientive (Patient- suppressing/demoting) ??Una persona così straordinaria non si dimentica, neanche del dettaglio più insignificante. Intended: ‘Such a remarkable person cannot be forgotten, and does not forget even the most insignificant detail.’ (22) Coordination (different diatheses) a. Reflexive (duplex) vs. facilitative (Agent-suppressing/demoting) ??Il ragazzo si taglia da solo e la carne facilmente. Intended: ‘The boy cuts himself and the meat cuts easily.’ a. Agentless passive (Agent-suppressing/demoting) vs. depatientive suppressing/demoting) ??Questa storia e quella ragazza si sono dimenticate. Intended: ‘This story has been forgotten and that girl forgot something.’ 27 (Patient- As may be inferred from the previous examples, various types of textual and contextual cues can be exploited to bring out the intended meaning in an unambiguous way. The next section takes a closer look at the role of such cues in guiding the interpretation of multifunctional voice markers. 3.2. Solving the riddle: the role of context in underspecification and ambiguity As pointed out by Dik (1983:234-235), the application of valency-decreasing voice markers may lead to an interpretative “riddle”, centering on the question “how can a two-place relation be relevant to only a single entity?”. As stressed by Magni (2020:16), this potential indeterminacy engendered by the underspecification or ambiguity of voice markers must be solved by the language user in context, by resorting to a variety of linguistic and situational cues. The notion of ‘context’ is intended here in a broad sense, as encompassing at least the following dimensions: 1. the lexical meaning of the verb and its arguments; 2. the accompanying grammatical information; 3. additional co-textual and situational information. For both underspecification and ambiguity relations, one or more of these contextual clues may be relevant in deriving the appropriate interpretation. Let us see a few examples. 28 3.2.1. Lexical cues Oftentimes, the lexical meaning of the verb provides all the information needed to derive the correct interpretation of a multifunctional voice marker. Consider a marker which can express both reflexivity and reciprocity like German sich: a sentence like Sie trafen sich, with the verb treffen, ‘meet’, is most naturally interpreted as ‘They met’ (i.e. “they met each other”), whereas a reflexive interpretation ‘they met themselves’ is extremely unlikely. The semantics of the predicate may also play a role in terms of lexical restrictions (see van Lier & Messerschmidt 2022): such restrictions may sometimes be somewhat idiosyncratic, either intra- or crosslinguistically. In Italian, for instance, the verb assassinare ‘murder’, which denotes an illegal act of killing, may not enter reflexive or anticausative constructions; according to Mendikoetxea (1999), however, reflexivization – though not anticausativization – of the corresponding verb is possible in Spanish. While the latter restriction is semantically motivated (anticausativization is not available for predicates that lexicalize an agent- or manner-component, cf. Haspelmath 1987), this is not always the case: for instance, the Cavineña (Pano-Tacanan) suffix -tana only allows an antipassive reading with the verbs jipe- ‘approach’ and jaka- ‘abandon’, whereas with any other verb, even if semantically similar, only a passive interpretation is available (Guillaume 2008:265). Besides that of the verb, the lexical meaning of the argument(s) can also constrain the interpretation of voice markers. Compare the two Russian sentences in (23)-(24), both featuring the verb kusat’, ‘bite’: when the subject is sobaka ‘dog’, a patientless-antipassive interpretation suggests itself, whereas with the subject eda ‘food’, only a facilitative reading appears to make sense. 29 (23) Sobaka kusaet-sja. dog(F).NOM.SG bite.PRS.3SG-DEOBJ ‘A dog bites.’ (Haspelmath 2003:224) (24) Tak kafe vyglyadit snaruzhi: […] A thus café look.PRS.3SG outside vnutri […] tak and thus inside A vot eda kusayet-sja s and PTC food(F).NOM.SG bite.PRS.3.SG-ANTP with udovol'stviyem. pleasure(F).INS.SG ‘This is how the cafe looks from the outside: […] And this is how it looks inside: […] Anyway, the food is quite good.’ (Lit. “the food bites-sja with pleasure”) (https://zen.yandex.ru/media/psychologoped/vaskin-rabbit61dc5e1690bae1348129d40f) 3.2.2. Grammatical cues A first straightforward example of how grammatical information can determine the interpretation of voice markers concerns the reciprocal diathesis: with whatever lexical verb, and in whatever context, singular number is presumably sufficient to block a reciprocal interpretation. While pronominal reciprocals are possible in some languages with formally singular collective nouns, we are not aware of any language in which such nouns allow for a reciprocal reading of a multifunctional voice marker 30 Number also plays an important role in disambiguating between an agentless-passive and a subjectless-impersonal interpretation. Returning to the opposition, introduced in 2.1.2, between Portuguese vendem-se casas and vende-se casas (both roughly translatable as ‘houses are sold’ or ‘one sells houses’), plural agreement unambiguously indicates that the former is an agentless passive, whereas vende-se casas, due to lack of agreement, can only be understood as a subjectless impersonal. Where agreement is not sufficient, word order may also supply relevant contextual cues. Given that subject status is typically associated with high topicality and expressed in preverbal position in Portuguese, an announcement like vende-se casa, hanging on a wall or published in a newspaper (where the house for sale has not been previously introduced and is therefore not topical) will trigger a non-promotional reading – i.e. casa will be understood as the direct object of vender. The other way round, if the single argument occurs in the position typically reserved to the subject, an agentless passive interpretation is strongly favored. Accordingly, nonpromotional constructions do not normally license a preverbal argument in (European) Portuguese, as illustrated in (25): (25) Estes livros this.M.PL book(M).PL não se podem NEG IMPRS can.IND.PRS.3PL can.IND.PRS.3SG levar para casa. bring.INF to house(F).SG ‘One cannot take these books home.’ 31 / *pode 3.2.3. Co-textual and situational cues When neither the lexical semantics of the predicate and its argument(s) nor the associated grammatical information are enough to guide the interpretive process, further lexical and/or extra-linguistic information has to be called upon. Let us return to German sich: out of context, a sentence like (26), with the verb hassen ‘hate’, allows for both a reflexive and a reciprocal reading: (26) Sie hassen sich. they hate REFL/RECP (i) ‘They hate themselves.’ (ii) ‘They hate each other.’ (Heine & Miyashita 2008:207) With such verbs, an adverb like gegenseitig, ‘mutually’, can be added to disambiguate between the two interpretations, ruling out a reflexive one: Sie hassen sich gegenseitig can only mean ‘They hate each other’ (Heine & Miyashita 2008:207; see also Nedjalkov 2007:164). In a similar vein, studies dedicated to disentangling the anticausative form the passive diathesis have pointed out that only the former licenses the occurrence of adverbial phrases meaning ‘by oneself’ and/or NPs expressing Cause or Force, e.g. Italian la finestra si è rotta da sola per il vento ‘the window broke by itself because of the wind’ (Schäfer 2008:Ch. 4). Even with a semantically highly transitive verb like ‘kill’, which is typically used in the literature to illustrate the reflexive use of multifunctional voice markers, the co-occurring lexical material may force out an alternative interpretation. While out of context French Il s’est tué is 32 prone to be interpreted as ‘He killed himself’, in Il s’est tué en tombant du toit ‘He got killed falling from the roof’, the ensuing adverbial clause makes it clear that the intended reading is an anticausative one (similarly to Italian si uccisero in (21a) above). Finally, the riddle may be solved thanks to the extralinguistic context. If the Russian verbform kusaetsja in (23)-(24) occurred with a deictic pronoun instead of a lexical subject, the appropriate interpretation could only be derived by establishing, in context, what type of entity the pronoun deictically refers to. Similarly, in French, saying Il s’est tué while commenting on the discovery of a hanging body will license a different reading than when the same sentence is uttered as a comment about a car accident (again, reflexive in the former case, anticausative in the latter). Context also plays a crucial role in the diachronic development of linguistic expressions. The ways in which this role manifests itself in the diachrony of voice markers will be the focus of Section 4. 4. The diachrony of voice markers Research on the grammaticalization of voice markers has shown that these originate from a wide array of sources and following a variety of developmental paths. Common lexical sources include verbs like ‘eat’ and ‘go’ for passive markers, verbs like ‘fall’ and ‘become’ for anticausatives, nouns denoting body parts for reflexives and generic nouns like ‘thing, stuff’ for antipassives (Kuteva et al. 2019, Zúñiga & Kittilä 2019:Ch. 8; Grestenberger & Kamil forthc.). More interesting, to our aims, is however the secondary grammaticalization of voice markers, that is, 33 the process whereby already grammatical constructions acquire new grammatical meanings (Givón 1991; see Breban 2014, 2015 for a recent reassessment). Secondary grammaticalization provides the diachronic explanation for the syncretism patterns surveyed in Section 2.2: voice markers originally restricted to one single diathesis are extended to the expression of other ones, while at the same time also retaining their original function, thus bringing about a situation of multifunctionality. While this is pretty much how new grammatical meanings are constantly created in all areas of the grammar, we have already remarked that the most striking aspect of the diachronic development of voice markers is the lack of unidirectionality in their secondary grammaticalization, unlike what seems to be the case for the overwhelming majority of grammatical categories. In Section 4.1 we review the evidence concerning multidirectionality in the voice domain, while in Section 4.2 we discuss the role of context, and of underspecification and ambiguity in the diachrony of voice. Section 4.3 links multidirectionality in the diachrony of voice to the nature of voice markers as interface operators. 4.1. Survey of diachronic changes in the voice domain Studies on the diachrony of voice markers have repeatedly pointed out that the reflexive function is the prime source for all other non-voice-increasing diatheses, so much so that this pathway is largely held to be unidirectional (e.g. Kemmer 1993, Haspelmath 1990). Drawing on the existing literature on the grammaticalization of voice, one can indeed detect two often-recurring diachronic pathways that originate in reflexive voice markers, as illustrated in Figure 3. On the one hand, reflexive markers may extend to the encoding of 34 reciprocity (Heine & Miyashita 2008), and reciprocal markers may subsequently acquire antipassive functions (Sansò 2017). Concerning the lower pathway, reflexives may develop into anticausative markers and these into passives, possibly via a facilitative stage (Haspelmath 1990). Finally, agentless and non-promotional passives are commonly believed to represent a further evolution from canonical passives (cf. Giacalone Ramat & Sansò 2011 on Italian). Figure 3: The secondary grammaticalization of reflexives (adapted from Inglese 2023:6) While the validity of the grammaticalization pathways in Figure 3 cannot be undermined, their unidirectionality can be challenged. In fact, there exist developments that either by-pass some of the stages in Figure 3 or follow alternative pathways, and even a few ones that contradict the directionalities of those in the figure. For reasons of space, we cannot discuss this evidence in detail, and refer to the data gathered in Bahrt (2021:Ch. 7) and Inglese (2023). First, antipassives may develop directly from reflexives without an intervening reciprocal stage (Holvoet 2020:65-67 on Baltic) and, likewise, anticausatives may develop into passives without going through a facilitative stage (Ahn & Yap 2017:457-458 on Korean). Other developments not captured in Figure 3 include, for example, the shift from reciprocal to anticausative, proposed for a number of Bantu languages by Dom et al. (2016). Concerning counter-directional changes, the following have been discussed in the literature: from 35 anticausative to reflexive and reciprocal (e.g. Authier 2012 on East-Caucasian, Inglese 2020 on Hittite), from passive to reflexive (e.g. Langacker & Munro 1975:803 on Uto-Aztecan), from passive to anticausative (e.g. Kulikov 2010 on Vedic Sanskrit), and from reciprocal to reflexive (this is the case of the Tuvan suffix -ž- in (7), see Gandon 2018; see also Waters 1989:88-89, Alpher et al. 2003:343, Evans et al. 2007 and Moyse-Faurie 2008 on Australian and Oceanic languages and Angster forthc.: 201-202 on Breton). To sum up, while reflexives are an important source for other non-valency-increasing voice functions, other possible changes, including a number of bidirectional ones, must also be reckoned with. Specifically, the view that reflexives always provide the source meaning is unwarranted, as reflexivity may also be the endpoint in the development of voice markers. Taking this evidence into consideration leads to a substantial modification of the schema in Figure 3, as shown in Figure 4 (dashed arrows stand for changes for which the empirical evidence is less robust). Figure 4: The secondary grammaticalization of voice markers, revised (adapted from Bahrt 2021:230) 36 4.2. The role of context in the grammaticalization of voice markers In Section 4.1, we have surveyed possible developments in the domain of voice, by merely listing attested source-target developments (in the spirit of Kuteva et al. 2019), i.e. without discussing how such historical changes take place in practice. Given the limited scope of this paper, we cannot present all these changes in detail. Instead, the point that we would like to stress is that these changes are possible because of two mechanisms, that is, extension and reinterpretation, which in turn we suspect to be connected to underspecification and ambiguity (on the role of ambiguity in language change see also Magni 2016:24-25, 2020:34-41, this volume).10 Extension takes place when speakers start applying a given voice marker to new (classes of) verbs, which are semantically similar to the ones originally used with the markers, without necessarily changing the underlying macro-diathesis (thus engendering underspecification). By contrast, reinterpretation takes place when, in specific contexts, voice markers may activate two readings belonging to different diatheses, that is, when ambiguity arises. Let us illustrate these points by considering the development of reflexives into anticausatives, as discussed by i.a. Haspelmath (1987, 1990), Heine (2002) and Holvoet (2020:Ch. 6). The starting point of this process are voice markers that only signal reflexivity, that is, coreference between the (volitional) Agent and the Patient of typically highly agentive verbs, as in (27a). From these prototypical reflexive contexts, speakers may extend reflexive marking to situations that are semantically close to reflexives but lack a clear differentiation between the Agent and Patient role (cf. Kemmer 1993), that is, to autocausative contexts, as in (27b) (see 10 We prefer the terms extension and reinterpretation over analogy and reanalysis because the latter have been extensively used in historical linguistics with at times conflicting meanings. 37 Section 2.1.3; Holvoet 2020:Ch.2 labels the verbs typical of such contexts naturally reflexive verbs). Crucially, provided the appropriate context, the subject of a verb like Italian alzarsi, ‘rise’, may be interpreted either as a volitional one, as in (27b), or as a non-volitional one, as in (27c), where alzarsi arguably indicates a non-controlled physical response to a visual stimulus. One can therefore say that alzarsi is underspecified with respect to the [±control] feature, hence with respect to the semantic role of its subject. Subsequently, from contexts such as (27c), where the volitional restriction on the subject is dropped, the same type of verb may be extended to subjects that cannot possibly be construed as volitional ones, that is, to Patient-only subjects, as in (27d). Once this reading becomes available, the marker can be extended to anticausative contexts proper, that is, to spontaneous change-of-state events that cannot possibly be construed as featuring an Agent, not even an external one, as in (27e). At this point, the anticausative use of the marker can be extended to other classes of verbs, as in (27f). In this development, it is underspecification in the bridging contexts (Heine 2002) (27b) and (27c) that allows the activation of both the source and the target meaning. It is only once full reinterpretation has taken place, that is, once the marker can be used in both (27a) and (27e-f) that a situation of true ambiguity can be said to have emerged. (27) Stage I: reflexive (A + volition) a. L’uomo si guarda allo specchio. ‘The man looks at himself in the mirror.’ Stage II: autocausative (A ± volition) b. L’uomo si alzò per andare in cucina. ‘The man stood up to go to the kitchen.’ 38 c. Avendo visto un ragno, l’uomo si alzò di scatto. ‘Having seen a spider, the man sprang up.’ Stage III: anticausative (P – volition) d. Il sipario si alza. ‘The stage curtain rises (is lifted).’ e. La marea si alza. ‘The tide rises.’ f. La neve si scioglie. ‘The snow melts.’ That in bridging contexts the relation between the source and the target meaning is indeed one of underspecification can again be shown by applying the zeugma test. Consider (28), where the first coordinated verb (alzarsi ‘get up’) triggers an autocausative reading with a volitional agent and the second one (lavarsi ‘wash oneself’) a purely reflexive interpretation: (28) Dopo essersi alzato e lavato, andò al lavoro come tutti i giorni. ‘After getting up and washing himself, he went to work like every day.’ If the reflexive and the volitional-autocausative interpretation can occur side by side without generating a contradiction, this means that both must belong to the same type of diathesis – namely, the duplex cluster. In (29), however, the first clause suggests a non-volitional autocausative reading: this can felicitously be coordinated with an elliptical clause with inanimate subject, which only allows for an anticausative interpretation. If (29) is acceptable 39 even with ellipsis of the verb in the second clause, this can only mean that the non-volitional autocausative reading stands in a relation of underspecification to the anticausative one, that is, both belong to the Agent-suppressing cluster. (29) Avendo visto un ragno, l’uomo si alzò di scatto – e la sedia con lui. ‘Having seen a spider, the main sprang up – and the chair (was lifted) with him.’ By contrast, if we try and coordinate a clause which only admits a reflexive reading with one that can only be interpreted as an anticausative, the zeugma test fails, indicating that the siconstruction has now become truly polysemous. This is illustrated in (30): (30) ??L’uomo si punse intenzionalmente con la siringa ed infettò. Intended: ‘The man intentionally pricked himself with the syringe and got infected.’ Interestingly, autocausatives not only play a crucial role in the shift from reflexives to anticausatives, but have also been argued to constitute the bridging context for the reverse development, from anticausative to reflexive (Inglese 2020:235-237), and from reflexive to antipassive (Holvoet 2020:65-67). This central role of autocausatives in the network of diachronic changes undergone by voice markers ties in very well with the proposal that these verbs share properties of more than one diathesis cluster (see Section 2.1.3). While this makes it difficult to synchronically ascribe them to a single cluster, it also makes them perfect candidates as contexts in which the various diatheses may evolve into one another. Crucially, since 40 underspecification in the interpretation of autocausatives works both ways, this partly explains why these verb types are a prime context in which bidirectional changes may take place. This brief overview of how reflexives may develop into anticausatives suggests that underspecification plays a crucial role in activating potential new readings of voice markers, and once the reinterpretation of these as expressing a new diathesis type is achieved, ambiguity steps in. More corpus-based research is needed to fully explore the validity of this scenario and to better disentangle the role of underspecification vs. ambiguity in the development of voice markers and other grammatical categories. 4.3. Explaining bidirectionality: Voice markers as interface operators As a measure of the exceptionality of bidirectional diachronic connections, suffice it to mention that of the 529 paths of grammaticalizations surveyed in in Kuteva et al. (2019), only fifteen are reported, or at least hypothesized to be bidirectional. Somewhat surprisingly, none of these involves voice markers. However, we have seen that the evidence for bidirectional diachronic changes in this domain is indeed solid. Now, if bidirectional grammaticalization channels are on the whole so rare, how come this phenomenon, though certainly not predominant, is not at all exceptional when it comes to the particular grammatical category of voice? To answer this question, it does not suffice to acknowledge the existence of contexts which, given two possible functions of voice markers, may allow for a pragmatically plausible reinterpretation in either direction. If this was the only reason for bidirectional developments, then such developments could reasonably be expected to 41 be much more widespread and to involve potentially any area of the grammar. Consider for instance (31): (31) The exam is tomorrow and I still haven’t read the last chapter. I will study all night. Despite its historical origin as a verb of volition, the auxiliary will is nowadays fully grammaticalized as a future marker. Moreover, in a context such as (31) it can easily be understood as implying a modal overtone of necessity. Since future markers often occur in this type of contexts, it is not far-fetched to imagine that such context-specific uses could in principle function as a bridge from future to necessity, thus enabling a reinterpretation process that would eventually lead to the full conventionalization of the latter meaning. However, while it is not hard to find contexts of the type of (31), in which future-marked forms may also imply a deontic meaning, a full-scale FUTURE > NECESSITY shift is to the best of our knowledge unattested; what is attested for several languages is, on the contrary, a development from necessity to future (e.g. Fleischman 1982, Andersen 2006 and Hilpert 2008 on Indo-European and Metslang 2016 on Uralic). Likewise, given a context like (32), where a specific time interval is explicitly indicated, the past habitual marker used to could easily be reinterpreted as a marker of progressive aspect: (32) In those days I used to take Yoga lessons. Again, however, habitual markers have not been found to develop into progressive markers as far as we can tell, whereas the opposite path is well documented (Bybee et al. 1994). 42 The existence of potential bridging contexts for bidirectional developments, in sum, is not a sufficient explanation for the relative frequency of such changes in the field of voice markers. To make sense of this uncommon diachronic tendency, it will be useful to reflect, on the one hand, on the reason why secondary grammaticalization is so largely unidirectional; and, on the other, on the very nature and functional import of the grammatical category of voice. The principles responsible for the predominant unidirectionality of functional change have been described in different, but to a large extent mutually compatible terms in the literature. At least the following generalizations deserve to be mentioned here: a. The so-called localist hypothesis (e.g. Lyons 1967, Anderson 1971, Jackendoff 1972, Heine et al. 1991), i.e. the idea that spatial notions are conceptually more basic than notions pertaining to other domains, and are therefore commonly exploited as structural templates for the expression of other types of meanings. b. Lehmann’s (1982, 2015) notion of semantic “bleaching” (or desemanticization), which encompasses the predictions that grammaticalizing elements will over time tend to acquire increasingly abstract, general and context-dependent meanings. c. The concept of subjectification, which in turn has been framed in different terms by different authors, but, generally speaking, can be summarized as a prediction that the meaning of grammatical constructions will become more and more speaker-bound as grammaticalization proceeds (e.g. Traugott 1982, 2010, Langacker 1990, 1998, Traugott & Dasher 2002, Narrog 2012). d. The associated notion of intersubjectification, according to which highly grammaticalized constructions typically “come to express grounding in the relationship between speaker/writer 43 and addressee/reader” (Traugott & Dasher 2002:6; see also Traugott 2010, Narrog 2012, Brems et al. 2014). In most accounts, this includes the acquisition of functions related to social deixis, face and discourse structure. e. An increase in semantic and/or pragmatic scope along a layered hierarchy of functional categories at whose top are interpersonal functions such as those mentioned in (d), then subjective ones like epistemic modality, and further down the hierarchy more and more objective specifications such as tense, root modality, event quantification, aspect, and so forth (Hengeveld 1989, 2011, 2017, Nuyts 2001, 2004, Narrog 2012). As Harder & Boye (2012:15) put it “[t]he principle underlying layered semantic structure and the associated scope hierarchy is [that] higher layers are conceptually dependent on lower layers; lower layers are functionally dependent on higher layers”. If we now consider the various non-valency-increasing diatheses discussed above in terms of the notions in (a)-(e), it seems to us that these are not easily ranked in terms of greater or lesser bleaching, (inter)subjectivity, etc. For instance, is the notion of agent suppression or demotion more or less general in meaning than that of patient suppression or demotion? And is passivization more or less (inter)subjective than antipassivization? In this regard, it should be stressed that the passive and antipassive are usually argued to be primarily triggered by matters of topicality and textual coherence (see especially Givón 2001) and may therefore be claimed to be more intersubjective than the other diatheses. In our view, the choice to, say, suppress or demote an Agent or Patient is equally pragmatically motivated: for instance, if one chooses to say The door opened instead of The wind opened the door or Something opened the door, this is because the involvement of a (known or unknown) Agent is deemed irrelevant from a 44 communicative viewpoint (see Langacker 2006:129-130 and Keizer 2014:415 on the so-called English middle voice). This difficulty of ranking the various voice oppositions in terms of the generalizations listed above not only applies across different diathesis-shift clusters, but also within each individual cluster. For instance, are reflexives and non-promotional passives more or less abstract, (inter)subjective, etc. than, respectively, reciprocals and subjectless impersonals? Also when approached from the perspective of the layered models of clause structure mentioned in (e), it is far from obvious whether any particular diathesis or diathesis-cluster may be said to have wider or narrower semantic scope than any other. Which takes us to the second issue mentioned above, that is, the very functional import of the grammatical category of voice. Since voice essentially consists in the marking of various types of mappings between semantic and syntactic roles, it is after all not surprising that the various functions of these markers cannot be assigned to any specific layer within a hierarchy of semantic categories; rather, it seems reasonable to regard diathesis shifts as operations that take place at the interface between semantics and syntax. What we would like to suggest, following Giomi (2020, 2023), is that this particular status of voice markers as interface operators is precisely the reason why the various notions in (a)-(e) above do not play a role in constraining the diachronic development of these constructions. Working within the layered model of Functional Discourse Grammar (Hengeveld & Mackenzie 2008), Giomi (2020, 2023) reviews hundreds of grammaticalization paths discussed in the diachronic literature, finding that the overwhelming majority of these appear to be indeed invariably unidirectional. This is explained as resulting from a constraint that changes affecting the meaning of linguistic expressions must follow an upward direction with respect to the 45 hierarchy of semantic and pragmatic categories endorsed by Functional Discourse Grammar, where pragmatics (captured at the Interpersonal Level of the model) governs semantics (captured at the Representational Level) and both govern morphosyntax and phonology, as illustrated in Figure 5. Figure 5: General layout of Functional Discourse Grammar (Hengeveld & Mackenzie 2008:13) Within both the Interpersonal and the Representational Level, pragmatic and semantic categories are organized hierarchically in such a way that lower layers are recursively embedded within higher and higher ones. For instance, at the Representational Level, argument structure is described at the layer of Configurational Properties (i.e. the abstract “situational concept” of the predication, see Hengeveld 2011), which captures such event-internal grammatical qualifications as phasal aspect (e.g. progressive) or participant-oriented modality (e.g. intention, ability, etc.); 46 the Configurational Property forms the head of the layer of States-of-Affairs, where eventexternal qualifications such as tense become relevant, and which in turn specifies the head of the Propositional Content (i.e. the locus for subjective grammatical qualifications like epistemic modality or inferential evidentiality). Thus, as a grammatical morpheme such as for instance English will grammaticalized from an intention to a future tense marker and then as a marker of epistemic modality, its scope progressively increased along the hierarchy of semantic categories. Finally, in Present Day English, will has acquired discourse-oriented uses such as mitigation (e.g. A: How much will it be? B: That’ll be dollar forty, see Halliday & Hasan 1985: 69), thus moving up from the Representational to the Interpersonal Level. As regards the directionality of grammaticalization, Giomi (2023) finds that the few grammaticalization channels that are not subject to strict directionality constraints all involve source and target meanings that belong to the same layer of the aforementioned semanticopragmatic scope hierarchy (Giomi 2023:138), or source/target pairs in which at least one of the two categories operates at the interface between two different levels of the grammar, and therefore is simply not part of that hierarchy of contentful functional categories, which pertains to the Representational and the Interpersonal level only. Relevant examples are nominalizers, complementizers, participle, infinitive and attributive markers and copulas (which all serve to adapt a given constituent, be it a single word, a phrase or a clause, to a semantic and syntactic slot in which it could not occur otherwise), dummy stem extenders (which serve purely phonological purposes) and indeed valency-increasing or decreasing voice markers (Giomi 2023:245-246,250-252,296). Assigning voice markers to the class of semantics-syntax interface operators is not only in accordance with familiar definitions of grammatical voice (see Section 2.1) but, most 47 importantly for the purpose of this paper, provides a straightforward explanation for the lack of unidirectionality in the development of these markers. In a nutshell, if voice is not strictly speaking a contentful grammatical category, in the sense that it does not express any specific conceptual notion but serves the merely structural function of signaling how semantic valency is mapped onto syntactic valency, then there is no reason why diachronic changes in this domain of the grammar should be governed by principles which, the other way round, are strictly conceptual in nature like those of semantic bleaching or (inter)subjectivity. It is for this reason that, when it comes to the development of voice markers, such principles do not impose any particular restriction on the types of contexts that may serve as bridges between two different interpretations of a single marker. 5. Conclusions In this chapter we have proposed an account of non-valency-increasing voice markers that attempts to bridge the gap between the multifunctionality of these constructions and the lack of unidirectionality in their diachronic development. Our proposal builds upon two main notional cores: a typology of diathesis shifts which exploits the distinction between underspecification and ambiguity and a characterization of voice as a grammatical category that operates at the interface between semantics and syntax. As regards the former, we have suggested that non-valency-increasing diatheses may be grouped into six clusters, each of which is characterized by a specific type of mapping between semantic and syntactic argument-roles, defining and delimiting as many distinct regions of 48 functional space. This means that the relation between different diathesis clusters is one of ambiguity, whereas that between the individual diatheses in each cluster is one of underspecification. This distinction was then brought into connection with the context-induced reinterpretation mechanisms that preside over the diachronic development of voice markers, in such a way that the kind of functional indeterminacy encountered in bridging contexts is understood in terms of underspecification, whereas true polysemy and ambiguity are only brought about by the full conventionalization of new grammatical functions. Finally, the lack of strict unidirectionality in these developments has been explained as a consequence of the status of voice markers as interface operators, the latter being adduced as the reason why oft-invoked cognitive/conceptual principles like semantic bleaching and (inter)subjectification fail to impose any particular constraint on the types of contexts that may enact the reinterpretation of voice markers as markers of new grammatical diatheses. References Ahn, Mikyung, and Foong H. Yap. 2017. “From Middle to Passive: A Diachronic Analysis of Korean -eci Constructions”. Diachronica 34 (4): 437–469. Alpher, Barry, Nicholas Evans and Mark Harvey. 2003. “Proto-Gunwinyguan Verb Suffixes”. In The Non-Pama-Nyungan Languages of Northern Australia: Comparative Studies of the Continent’s Most Linguistically Complex Region, ed. by Nicholas Evans, 305–352. Canberra: Australian National University. 49 Andersen, Henning. 2006. “Periphrastic Futures in Slavic: Divergence and Convergence”. In Change in Verbal Systems: Issues in Explanation, ed. by Kerstin Eksell, and Thora Vinther, 9–45. Bern: Peter Lang. Anderson, John M. 1971. The Grammar of Case: Towards a Localistic Theory. London: Cambridge University Press. Angster, Marco. Forthcoming. At the boundaries of word-formation. Contrastive coreference in 30 European languages. Zadar: University of Zadar. Authier, Gilles. 2012. “The Detransitive Voice in Kryz”. In Ergativity, Valency and Voice, ed. by Gilles Authier, and Katharina Haude, 133–164. Berlin: de Gruyter. Babby, Leonard H. 1999. “Voice and Diathesis in Slavic”. Paper presented at Comparative Slavic Morphosyntax (Bloomington, IN, 5–7 June). Bahrt, Nicklas N. 2021. Voice Syncretism. Berlin: Language Science Press. Bickel, Balthasar. 2011. “Grammatical Relations Typology”. In The Oxford Handbook of Language Typology, ed. by Jae Jung Song, 399–444. Oxford: Oxford University Press. Boeder, Winfried. 1969. “Uber die Versionen des georgischen Verbs”. Folia Linguistica 2: 82– 152. Breban, Tine. 2014. “What Is Secondary Grammaticalization? Trying to See the Wood for the Trees in a Confusion of Interpretations”. Folia Linguistica 48 (2): 469–502. Breban, Tine. 2015. “Refining Secondary Grammaticalization by Looking at Subprocesses of Change”. Language Sciences 47: 161–171. Brems Lieselotte, Lobke Ghesquière, and Freek Van de Velde (eds.). 2014. Intersubjectivity and Intersubjectification in Grammar and Discourse: Theoretical and Descriptive Advances. Amsterdam: John Benjamins, 50 Bybee, Joan L., Revere D. Perkins, and William Pagliuca. 1994. The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. Chicago: University of Chicago Press. Comrie, Bernard. 1989. Language Universals and Linguistic Typology: Syntax and Morphology (2nd edition). Oxford: Blackwell. Blackwell. Creissels, Denis. 2006. Syntaxe générale, une introduction typologique. Paris: Hermès. Creissels, Denis. Forthcoming. Transitivity, valency, and voice. Oxford: Oxford University Press. Croft, William. 2003. Typology and Universals. Cambridge: Cambridge University Press. Dik, Simon C. 1983. “On the Status of Verbal Reflexives”. In Problems in Syntax, ed. by Liliane Tasmowski, and Dominique Willems, 231–256. New York: Plenum. Dom, Sebastian, Leonid Kulikov, and Koen Bostoen. 2016. The Middle as a Voice Category in Bantu: Setting the Stage for Further Research”. Lingua Posnaniensis 58 (2): 129–149. Evans, Nicholas, Alice Gaby, and Rachel Nordlinger. 2007. “Valency Mismatches and the Coding of Reciprocity in Australian Languages”. Linguistic Typology 11: 543–599. Fleischman, Suzanne. 1982. The Future in Thought and Language: Evidence from Romance. Cambridge: Cambridge University Press. Gandon, Ophelie. 2018. “The Verbal Reciprocal Suffix in Turkic Languages and the Development of its Different Values”. In The Rouen Meeting: Studies on Turkic Structures and Language Contacts, ed. by Mehmet A. Akıncı, and Kutlay Yagmur, 29–43. Wiesbaden: Harrassowitz. Geniušienė, Emma. 1987. The Typology of Reflexives. Berlin: de Gruyter. 51 Giacalone Ramat, Anna, and Andrea Sanso. 2011. “From Passive to Impersonal: A Case Study from Italian and its Implications”. In Impersonal Constructions: A cross-linguistic Perspective, ed. by Andrej Malchukov, and Anna Siewierska, 189–228. Amsterdam: John Benjamins. Giomi, Riccardo. 2020. Shifting Structures, Contexts and Meanings: A Functional Discourse Grammar Account of Grammaticalization. University of Lisbon, PhD Dissertation. Giomi, Riccardo. 2023. A Functional Discourse Grammar Theory of Grammaticalization, Volume I: Functional Change. Leiden: Brill. Givón, Talmy. 1991. “The Evolution of Dependent Clause Morphosyntax in Biblical Hebrew”. In Approaches to Grammaticalization, Volume II: Focus on Types of Grammatical Markers, ed. by Elizabeth C. Traugott, and Bernd Heine, 257–310. Amsterdam: John Benjamins. Givón, Talmy. 2001. Syntax: An introduction. Amsterdam: John Benjamins. Grestenberger, Laura, and Iris Kamil. Forthcoming. “Voice alternations in diachrony. In Wiley Blackwell Companion to Diachronic Linguistics, ed. By A. Ledgeway, E. Aldridge, A. Breitbarth, K. É. Kiss, J. Salmons & A. Simonenko. Hoboken: Wiley-Blackwell. Guentchéva, Zlatka, and Nicole Rivière. 2007. “Reciprocal and reflexive Constructions in French”. In Typology of Reciprocal Constructions, Volume II, ed. by Vladimir P. Nedjalkov, 561–608. Amsterdam: John Benjamins. Guillaume, Antoine. 2008. A grammar of Cavineña. Berlin: de Gruyter. Halliday, M.A.K, and Ruqaiya Hasan. 1985. An Introduction To Functional Grammar. London: Arnold 52 Harder, Peter, and Kasper Boye. 2012. “Grammaticalization and (Inter)Subjectification”. In Grammaticalization and (Inter)Subjectification, ed. by Johan van der Auwera, and Jan Nuyts, 9–20. Brussels: Koninklijke Vlaamse Academie van Wetenschappen en Kunsten. Haspelmath, Martin. 1987. “Transitivity Alternations of the Anticausative Type”. Köln: Institut für Sprachwissenschaft. Haspelmath, Martin. 1990. “The Grammaticalization of Passive Morphology”. Studies in Language 14 (1): 25–72. Haspelmath, Martin. 2003. “The Geometry of Grammatical Meaning: Semantic Maps and Crosslinguistic Comparison”. In The New Psychology of Language: Cognitive and Functional Approaches to Language Structure, Volume II, ed. by Michael Tomasello, 211– 242. Mahwah: Erlbaum. Haspelmath, Martin. 2004. “On Directionality in Language Change with Particular Reference to Grammaticalization”. In Up and Down the Cline: The Nature of Grammaticalization, ed. by Olga Fischer, Muriel Norde & Harry Perridon, 17–44. Amsterdam: John Benjamins. Haspelmath, Martin, and Thomas Müller–Bardey. 2008. “Valency Change”. In Morphology: A Handbook on Inflection and Word Formation. Volume II, ed. by Geert E. Booij, Christian Lehmann, Joachim Mugdan, and Stavros Skopeteas, 1130–1145. Berlin: de Gruyter. Heine, Bernd. 2002. “On the Role of Context in Grammaticalization”. In New Reflections on Grammaticalization, ed. by Ilse Wischer & Gabriele Diewald, 83–101. Amsterdam: John Benjamins. Heine, Bernd, Ulrike Claudi, and Friederike Hünnemeyer. 1991. Grammaticalization: A Conceptual Framework. Chicago: University of Chicago Press. 53 Heine, Bernd, and Hiroyuki Miyashita. 2008. “The Intersection between Reflexives and Reciprocals: A Grammaticalization Perspective”. In Reciprocals and Reflexives: Theoretical and Typological Explorations, ed. by Ekkehard König & Volker Gast, 169– 223. Berlin: de Gruyter. Hengeveld, Kees. 1989. “Layers and Operators in Functional Grammar”. Journal of Linguistics 25: 127–157. Hengeveld, Kees. 2011. “The Grammaticalization of Tense, Mood and Aspect”. In The Oxford Handbook of Grammaticalization, ed. by Heiko Narrog & Bernd Heine, 580–594. Oxford: Oxford University Press. Hengeveld, Kees. 2017. “A Hierarchical Approach to Grammaticalization”. In The Grammaticalization of Tense, Aspect, Modality and Evidentiality: A Functional Perspective, ed. by Kees Hengeveld, Heiko Narrog, and Hella Olbertz, 13–38. Berlin: de Gruyter. Hengeveld, Kees & J. Lachlan Mackenzie. 2008. Functional Discourse Grammar: A Typologically–Based Theory of Language Structure. Oxford: Oxford University Press. Hilpert, Martin. 2008a. Germanic Future Constructions: A Usage-Based Approach to Language Change. Amsterdam: John Benjamins. Holvoet, Axel. 2020. The Middle Voice in Baltic. Amsterdam: John Benjamins. Hopper, Paul J., and Sandra A. Thomposon. 1980. “Transitivity in Grammar and Discourse”. Language 56 (2): 251–299. Inglese, Guglielmo. 2020. The Hittite Middle Voice: Synchrony, Diachrony, Typology. Leiden: Brill. 54 Inglese, Guglielmo. 2022a. “Towards a Typology of Middle Voice Systems”. Linguistic Typology 26 (3): 489-531. Inglese, Guglielmo. 2022b. “How do middle voice markers and valency reducing constructions interact? Typological tendencies and diachronic considerations”. Folia Linguistica 56(2): 239–271. Inglese, Guglielmo. 2023. “The Rise of Middle Voice Systems: A Study in Diachronic Typology”. Diachronica 40 (2): 195–237. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge: MIT Press. Janic, Katarzyna, and Alena Witzlack-Makarevich. 2021. “The Multifaceted Nature of the Antipassive Construction”. In Antipassive, ed. by Katarzyna Janic, and Alena WitzlackMakarevich, 1–39. Amsterdam: John Benjamins. Keizer, Evelien. 2014. “Context and Cognition in Functional Discourse Grammar: What, Where and Why?”. Pragmatics 24 (2): 399–423. Kemmer, Suzanne. 1993. The middle Voice. Amsterdam: John Benjamins. Kennedy, Christopher. 2019. “Ambiguity and Vagueness: An Overview”. Semantics: Lexical Structures and Adjectives, ed. by Claudia Maienborn, Klaus von Heusinger, and Paul Portner, 236–271. Berlin: de Gruyter. Kulikov, Leonid. 2010. “Voice Typology”. In The Oxford Handbook of Linguistic Typology, ed. by Jae J. Song, 368–398. Oxford: Oxford University Press. Kuteva, Tania, Bernd Heine, Bo Hong, Haiping Long, Heiko Narrog, and Seongha Rhee. 2019. World Lexicon of Grammaticalization (2nd edition). Cambridge: Cambridge University Press. 55 Kuular, Klara. 2007. “Reciprocals, Sociatives, Comitatives, and Assistives in Tuvan”. In Reciprocal constructions, ed. by Vladimir P. Nedjalkov, 1163–1230. Amsterdam: John Benjamins. Lakoff, George 1970. “A Note on Vagueness and Ambiguity”. Linguistic Inquiry 1 (3): 357–359. Langacker Ronald W., and Munro Pamela. 1975. “Passives and Their Meaning”. Language, 51 (4): 789–830 Langacker, Ronald W. 1990. “Subjectification”. Cognitive Linguistics 1: 5–38. Langacker, Ronald W. 1998. “On Subjectification and Grammaticization”. In Discourse and Cognition: Bridging the Gap, ed. by Jean-Pierre Koenig, 71–89. Stanford: CSLI. Langacker, Ronald W. 2006. “Dimensions of Defocusing”. In Voice and Grammatical Relations, ed. by Tasaku Tsunoda, and Tarō Kageyama, 115-138. Amsterdam: John Benjamins. Lehmann, Christian. 1982. Thoughts on Grammaticalization: A Programmatic Sketch. Cologne: University of Cologne. Lehmann, Christian. 2015. Thoughts on Grammaticalization (3rd edition). Berlin: Language Science Press. Lier, Eva van, and Maria Messerschmidt. 2022. “Lexical Restrictions on Grammatical Relations in Voice and Valency Constructions: An Introduction”. STUF – Language Typology and Universals 75 (1): 1–20. Lyons, John. 1967. “A note on Possessive, Existential and Locative Sentences”. Foundations of Language 3: 390–396. Magni, Elisabetta. 2016. “Sette tipi di ambiguità nel mutamento linguistico”. In Problemi e prospettive della linguistica storica, ed. by Patrizia Cordin, and Alessandro Parenti, 13–34. Roma: Il Calamo. 56 Magni, Elisabetta. 2020. “Ambiguity and Uncertainty in the Synchrony and Diachrony of Language”. Special issue of Quaderni di Semantica 2020: 13–46. Magni, Elisabetta. This Volume. “The Role of Ambiguity and Vagueness in Language Change.” Mendikoetxea, Amaya. 1999. “Construcciones con se: Medias, pasivas e impersonales”. In Gramática descriptiva de la lengua Española, ed. by Ignacio Bosque, and Violeta Demonte, 1631–1722. Madrid: Editorial Espasa. Metslang, Helle. 2016. “Can a Language Be Forced? the Case of Estonian”. In Aspects of Grammaticalization: (Inter)Subjectification and Directionality, ed. by Daniel Olmen, Hubert Cuyckens, and Lobke Ghesquière, 281–310. Berlin: de Gruyter. Moyse-Faurie, Claire. 2008. “Constructions Expressing Middle, Reflexive and Reciprocal Situations in some Oceanic Languages”. In Reciprocals and Reflexives: Theoretical and Typological Explorations, ed. by Ekkehard König & Volker Gast, 105–168. Berlin: de Gruyter. Murphy, M. Lynne. 2010. Lexical Meaning. Cambridge: Cambridge University Press. Murray, Sarah E. 2008. “Reflexivity and reciprocity with(out) underspecification”. In Proceedings of Sinn und Bedeutung 12, ed. by A. Grønn, 455–469. Oslo: ILOS. Narrog, Heiko. 2012. Modality, Subjectivity, and Semantic Change: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Nedjalkov, Vladimir P. 2007. “Encoding of the Reciprocal Meaning”. In Reciprocal constructions, ed. by Vladimir P. Nedjalkov, 147–208. Amsterdam: John Benjamins. Nedjalkov, Vladimir P., and Georgij G. Silnitsky. 1973. “The Typology of Morphological and Lexical Causatives”. In Trends in Soviet Theoretical Linguistics, ed. by Ference Kiefer, 1– 32. Dordrecht: Reidel. 57 Neshcheret, Natalia, and Alena Witzlack-Makarevich. 2016. “Passives with Intransitive Verbs: Typology and Distribution”. Paper presented at the 46th Poznań Linguistic Meeting, (Poznań, 15–17 September). Nuyts, Jan. 2001. Epistemic Modality, Language, and Conceptualization: A CognitivePragmatic Perspective. Amsterdam: John Benjamins. Nuyts, Jan. 2004. “Remarks on Layering in a Cognitive-Functional Language Production Model”. A new Architecture for Functional Grammar, ed. by J. Lachlan Mackenzie, and María de los Ángeles Gómez González, 275–298. Berlin: de Gruyter. Oikonomou, Despina, and Artemis Alexiadou. 2022. “Voice Syncretism Crosslinguistically: The View from Minimalism”. Philosophies 7: 19 https://doi.org/10.3390/philosophies7010019 Palmieri, Giada. 2020. On the ambiguity between reflexivity and reciprocity in Italian. In Proceedings of the Fiftieth Annual Meeting of the North East Linguistic Society, ed by M. Asatryan, Y. Song and A. Whitma (eds.) 279-289. Amherst(MA): GLSA Publications. Peres, João Andrade, and Telmo Móia. 1995. Áreas Críticas da Língua Portuguesa (2nd edition). Lisbon: Caminho. Sansò, Andrea. 2017. “Where do Antipassive Constructions come from?: A Study in Diachronic Typology”. Diachronica 34 (2): 175–218. Schäfer, Florian Mathis. 2008. The Syntax of (Anti-)Causatives: External Arguments in Changeof-State Contexts. Amsterdam: John Benjamins. Shluinsky, Andrey. 2021. “Verb-verb complexes in Turkic languages: Interaction of Lexical and Delexicalized Verbs”. Verb-Verb Complexes in Asian Languages, ed. by. Taro Kageyama, Peter E. Hook, and Prashant Pardeshi, 397–429. Oxford: Oxford University Press. 58 Sobin, Nicholas J. 1985. “Case Assignment in Ukrainian Morphological Passive Constructions”. Linguistic Inquiry 16 (4): 649–662. Traugott, Elizabeth C. 1982. “From Propositional to Textual and Expressive Meanings: Some Semantic-Pragmatic Aspects of Grammaticalization”. In Perspectives on Historical Linguistics, ed. by Winfred P. Lehmann, and Yakov Malkiel, 245–271. Amsterdam: John Benjamins. Traugott, Elizabeth C. 2010. “(Inter)Subjectivity and (Inter)Subjectification: A Reassessment”. In Subjectification, Intersubjectification and Grammaticalization, ed. by Kristin Davidse, Lieven Vandelanotte, and Hubert Cuyckens, 29–74. Berlin: de Gruyter. Traugott, Elizabeth C., and Richard B. Dasher. 2002. Regularity in Semantic Change. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2005. Exploring The Syntax-Semantics Interface. Cambridge: Cambridge University Press. Vigus, Meagan. 2018. Antipassive constructions: Correlations of form and function across languages. Linguistic Typology 22 (3): 339–384. Wasow, Thomas. 2015. “Ambiguity Avoidance is Overrated”. In Ambiguity: Language and Communication, ed. by Susanne Winkler, 29–47. Berlin: de Gruyter. Waters, Bruce E. 1989. Djinang and Djinba: A Grammatical and Historical Perspective. Canberra: Australian National University. Zúñiga, Fernando, and Seppo Kittilä. 2019. Grammatical Voice. Cambridge: Cambridge University Press. Zwicky, Arnold, and Jerrold Sadock 1975. “Ambiguity Tests and How to Fail Them”. In Syntax and Semantics 4, ed. by John P. Kimball, 1–36. New York: Academic Press. 59