Advances in Research on Semantic Roles [1 ed.] 9789027266798, 9789027242761

Citation preview

tilä Zúñi t i K o o SeppFernand and

.) s d e ga (

B E N J A M I N S C U R R E N T TO P I C S

ch sear in Re s nces Role Adva mantic e on S



Advances in Research on Semantic Roles

Benjamins Current Topics issn 1874-0081 Special issues of established journals tend to circulate within the orbit of the subscribers of those journals. For the Benjamins Current Topics series a number of special issues of various journals have been selected containing salient topics of research with the aim of finding new audiences for topically interesting material, bringing such material to a wider readership in book format. For an overview of all books published in this series, please see http://benjamins.com/catalog/bct

Volume 88 Advances in Research on Semantic Roles Edited by Seppo Kittilä and Fernando Zúñiga These materials were previously published in Studies in Language 38:3 (2014).

Advances in Research on Semantic Roles Edited by

Seppo Kittilä University of Helsinki

Fernando Zúñiga University of Bern

John Benjamins Publishing Company Amsterdam / Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

doi 10.1075/bct.88 Cataloging-in-Publication Data available from Library of Congress: lccn 2016018278 (print) / 2016026351 (e-book) isbn 978 90 272 4276 1 (Hb) isbn 978 90 272 6679 8 (e-book)

© 2016 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Company · https://benjamins.com

Table of contents Introduction: Recent developments and open questions in the field of semantic roles Seppo Kittilä and Fernando Zúñiga

1

Identifying semantic role clusters and alignment types via microrole coexpression tendencies Iren Hartmann, Martin Haspelmath and Michael Cysouw

27

Semantic role clustering: An empirical assessment of semantic role types in non-default case assignment Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich Semantic roles and verbless constructions: A Finnish challenge for verbcentered approaches Katja Västi and Seppo Kittilä Benefaction proper and surrogation Fernando Zúñiga Exploring a diachronic (re)cycle of roles: The Dative complex from Latin to Romance Chiara Fedriani and Michele Prandi

51

79 109

133

Functive phrases in typological and diachronic perspective Denis Creissels

173

Language index

217

Subject index

219

Introduction

Recent developments and open questions in the field of semantic roles Seppo Kittilä and Fernando Zúñiga University of Helsinki / University of Bern

This introductory chapter briefly introduces a few milestones in the voluminous previous literature on semantic roles, and charts the territory in which the chapters of this volume aim to make a contribution. This territory is characterized by fairly disparate conceptualizations of semantic roles and their status in theories of grammar and the lexicon, as well as by diverse and probably complementary ways of deriving or identifying them based on linguistic data. Particular attention is given to the question of how selected roles appear to relate to each other, and we preliminarily address the issue of how roles, subroles, and role complexes are best thought of in general. Keywords: semantic role, subrole, role complex, lexical semantics

1. Preliminaries “In my view there is no construct as murky in any subdivision of linguistic theory as that of ‘thematic role’. Literally dozens have been proposed over the years, and nothing approaching a consensus has been achieved in terms of delineating the set that are needed for natural language semantics” (Newmeyer 2010: 689, emphasis in the original). This quote from Newmeyer can be seen as relevant to the chapters of this volume. The workshop (where all the chapters of this volume were presented),1 organized by the editors of this volume, aimed at making the theory of semantic roles more focused by discussing them from novel perspectives. The talks delivered at the workshop dealt with several different role complexes in which the differences 1.  The workshop took place at the University of Zurich in April, 2011. The chapters originally appeared in a special issue of Studies in Language (38.3) in 2014. We are grateful to Florian Matter for his valuable help with the indexes of this volume. doi 10.1075/bct.88.01kit 2016 © John Benjamins Publishing Company

2

Seppo Kittilä and Fernando Zúñiga

between semantic roles may be rather minimal, but a distinction may be based on other cues, such as the semantics of the participants instead of, for example, their case marking. Not all the talks appear in the present book, but we nevertheless hope that the ones that are included make a contribution to our understanding of semantic roles, although we do not wish to claim that a kind of consensus referred to by Newmeyer has been achieved. Semantic roles may be, and have been, defined in numerous ways depending on who has defined them and for what purpose. This is the topic of Section 2 of this introduction, where we present some previous, more or less established views of semantic roles. Moreover, the proposed definitions of semantic roles differ drastically in whether they see semantic roles as properties of nominal phrases, referents, verbs, events, or something else; we focus on these aspects in Section 3. Section 4 aims at shedding some light on whether different ‘senses’ of the same basic role should be seen as a single role or as a bundle of (closely related) roles. Finally, form and meaning contribute to our interpretation of semantic roles in different ways, which we discuss in Section 5. Section 6 is a brief summary of the central findings of the chapter and a survey of the chapters of the volume. 2. Previous studies on semantic roles The chapter ‘The case for Case’ by Charles Fillmore (1968) marks an important milestone for our current understanding of semantic roles. Case Grammar (henceforth CG) analyzes the surface syntactic structure of English sentences by studying deep cases (i.e. semantic roles in contemporary terminology), such as Agent, Object, Dative, Location, or Instrument, which are required by verbs. For instance, the verb give requires an Agent and Object (Patient), and a Dative. Verbs select a certain number of semantic roles which form their case frame, with any individual role able to occur only once per sentence. Unlike optional roles like Location, obligatory roles like Agent may not be omitted, at the risk of producing ungrammatical sentences (e.g. *gave the apples to Barry). Fillmore’s theory explicitly distinguishes between syntax and semantics, which means, for example, that the semantic roles of active and passive constructions are identical, whereas the formal manifestation of the roles is different in each construction. This idea is crucial to many of today’s functional-typological approaches to semantic roles, which attempt to be language-independent and thus make it possible to study semantic



Recent developments and open questions in the field of semantic roles

roles in radically different languages without supposing any kind of formal correlates of certain semantic roles, such as nominative case or preverbal position.2 Thematic relations have been acknowledged in formalist approaches to grammar since Gruber (1965), but they arguably have a varied status in this body of literature. Even though Chomskyan frameworks distinguish between semantic notions like thematic relations and syntactic notions like theta roles (e.g. Reinhart 2002), which are regarded as bundles of thematic relations associated with particular argument positions (e.g. Carnie 2002), several studies in this tradition have used both terms interchangeably. Relational Grammar (e.g. Perlmutter & Postal 1984), Government and Binding (e.g. Baker 1988), Lexical-Functional Syntax (e.g. Bresnan 2001), and other theories have used hierarchies of primitive thematic relations for particular formulations of mappings between the latter and argument positions, but some studies do without them altogether. Hale & Keyser (1993, 2001), for instance, claim that thematic roles and theta roles are derivative and non-essential. By a similar token, Jackendoff ’s (1983, 1990) work, as well as, more recently, Culicover & Jackendoff 2005, propose fairly sophisticated representations of semantic structure regularities, according to which thematic roles and theta roles are largely redundant. Role and Reference Grammar (henceforth RRG) distinguishes between predicate-specific semantic roles and predicate-class abstractions called thematic relations. For instance, in Frodo saw the ring, Frodo is the seer and the ring is the thing seen (semantic roles), but Frodo is also an experiencer and the ring a stimulus (thematic relations). Crucially, however, thematic relations are not primitives in this theory, but merely descriptive labels given to specific argument positions of particular semantic decompositions of predicates and their arguments. The following list illustrates how such thematic relations relate to individual predicates and their arguments according to RRG: (1)

Thematic relations continuum (Van Valin & LaPolla 1997: 127) a. 1st argument of DO (x, ): agent b. 1st argument of do′ (x, ): effector, consumer, creator, mover, etc. c. 1st argument of predicate′ (x, y): possessor, experiencer, wanter, location, etc.

2.  Later work by Fillmore (and others) on Frame Semantics and the FrameNet Project (see https://framenet.icsi.berkeley.edu/fndrupal/) distinguish between semantic roles and frame elements. While the latter merely correspond to event participants involved in semantic frames evoked by lexical units, semantic roles are their “linguistically motivated abstractions in that they pick out specifically those properties that tend to display the same behavior in morphosyntax” (Fried & Östman 2004: 42).

3

4

Seppo Kittilä and Fernando Zúñiga

d. 2nd argument of predicate′ (x, y): possessed, creation, theme, consumed, stimulus, desire, performance, etc. e. Argument of state predicate′ (x): patient, entity

The most recent proposal regarding the issues at hand we are aware of can be found in Croft (2012), which builds upon ideas already formulated by the same author in several studies over the last twenty-odd years and — in quite a similar spirit to RRG — uses semantic role labels as notional terms without theoretical significance. In Croft’s so-called force-dynamic approach,3 crucial reference is made to the causal structure of events, which most prominently feature initiators and endpoints, each of which can in turn be either physical or mental. In the English clause Sue broke the coconut for Greg with a hammer (Croft 2012: 206), the (physical) initiator (Sue) is the subject while the (physical) endpoint (the coconut) is the object, and there also is an antecedent oblique (a hammer) and a subsequent oblique (Greg) that correspond to an Instrument and a Beneficiary, respectively. The portion of the causal chain leading from Sue through the hammer to the coconut is said to be a causal segment profiled by the verb break in the so-called transitive argument construction just illustrated; the last segment, i.e. the one including Greg, is profiled by the preposition for. Most importantly, Croft’s approach does not resort to the notion of semantic roles at all, “either reified as separate semantic units or even defined as positions in argument structure” (p. 207): instead, there is a direct mapping between so-called semantic representation (i.e. the causal chain with its verbal profile) and argument roles (i.e. subject, object, and obliques). (See Croft 2012: Chapter 5 for more details on terminology and analysis.) Other functional approaches propose a reduction of semantic roles to other, more primitive semantic notions. Rozwadowska (1988, 1989) is a case in point: based on the three binary features [±change], [±sentient], and [±cause], she defines and distinguishes roles like Affected Agent, (prototypical) Agent, Experiencer, Instruments, Patients, Objects, Neutrals, etc. Analogously, Næss (2007) works with the binary features [±affectedness], [±volitionality], and [±instigation] and achieves more or less the same task. For example, in Rozwadowska the (canonical) Agent is [+sentient], [+cause], and [−change], while the typical Patient constitutes the opposite of this and thus has the features [−sentient], [−cause], and [+change]. In functional-typological linguistics (especially in Basic Linguistic Theory, see e.g. Dixon 2010), semantic roles are seen as language-independent descriptive and comparative tools that make it possible to study the formal expression of grammatical relations across languages. Role definitions are typically based on 3.  Croft’s work has elaborated notions proposed by both Leonard Talmy in the 1970s and 1980s and Ronald Langacker in the 1990s for the study of how purported real-world events are conceptualized and expressed in natural languages.



Recent developments and open questions in the field of semantic roles

semantic features such as affectedness, control, and volitionality. Roles are distinguished based on both semantic and formal cues; the latter include case or adposition marking and compatibility with certain adverbial expressions. The number of roles distinguished varies substantially, depending, for example, on whether differences in overt coding are deemed crucial to establishing a role, and on whether verbal semantics or syntax are at the center of attention. The number of roles distinguished in syntax-centered studies is rather low, because languages do not make formal distinctions among, say, 50 roles, but usually a maximum of 15–20 roles may be separated from each other solely through formal criteria. Examples of rather typical definitions of central semantic roles in functional-typological linguistics are provided by the following definitions of Proto-Agent (1) and ProtoPatient (2) by Dowty (1991) in terms of the logical entailments of these roles:4 (2)

Properties of the Proto-Agent (Dowty’s “Agent Proto-Role”) a. volitional involvement in the event or state b. sentience (and/or perception) with respect to the event or state c. causing an event or change of state in another participant d. movement (relative to the position of another participant) e. existence independent of the event or state

(3)

Properties of the Proto-Patient (Dowty’s “Patient Proto-Role”) a. is controlled (volitionally affected) by another participant b. is causally affected by another participant c. undergoes a change of state (e.g. is moved or physically manipulated by another participant) d. is the target of sentience of another participant e. existence depends on another participant or on the event or state

In Cognitive Construction Grammar (henceforth CxG), lexical semantics of nouns do not strictly define their roles, nor does verbal semantics directly determine the roles arguments take (Goldberg 1995, 2006). Rather than features of verbal semantics, semantic roles are considered constructional properties that encode event types basic to human experience, on occasion according to different alternative construals (Goldberg 1995: 39–40). According to this view, it is possible to use verbs in different constructions, which accord different semantic roles to their arguments, as well as other semantic and pragmatic properties. For example, the semantic role of the key is labeled as instrument in both John opened the gate with the key and The key opened the gate in Case Grammar, whereas in CxG the key can 4.  Dowty’s definitions are based on English only and therefore not necessarily language-independent, but they can also be seen as representative examples of how semantic roles may be defined in functional-typological linguistics.

5

6

Seppo Kittilä and Fernando Zúñiga

also be conceptualized metaphorically as an agent or force due to its appearance as the subject of the latter clause. A further distinction relevant to CxG is the one between argument roles and participant roles. The former roughly correspond to RRG’s thematic relations (e.g. Agent, Recipient, and Patient), whereas the latter are semantically more specific and correspond to RRG’s semantic roles (e.g. the argument role Agent can be divided into buyer, hander, and giver, depending on the semantics of the construction in question). 3. Where do semantic roles originate/come from? Predicates and the events they denote have been relevant to how semantic roles have been defined in the literature; if there is a verb present, it is usually most directly responsible for the semantic role assignment in the construction in question. For example, with the verbs break and paint we have an Agent and a Patient, while the verbs love and hate license an Experiencer and a Stimulus. The verb also sets certain limits for the arguments it can appear with, both semantically and formally (bearing in mind that formal (case) marking is often motivated by semantics). In many cases, we are dealing with finer nuances, with the verb kill, for example, the object usually refers to someone that is caused to die, which makes inanimate referents rather marginal (or at least figurative and arguably non-prototypical) with that verb, while not, however, affecting the basic patienthood of the direct object referent. In typical cases, the verb is responsible for the semantic roles, while the arguments (if present) specify the identity of the participants. The lexical semantics of arguments thus becomes less relevant whenever a verb is present. However, the verb does not always assign semantic roles directly, but there is some variation as well, as in John burnt the house vs. the lightning burnt the house, where John is a canonical Agent, while lightning is best seen as Force. Therefore, if arguments are (formally) left out, we only lose information about the exact identity of a participant (along with some potential changes in the roles), but in cases where verbs are omitted, the consequences for the reading of a clause are usually more dramatic. Moreover, different semantic roles can be assigned to arguments of different lexical semantics, even though these arguments show no formal coding distinction. For example, the English verb shoot may take different kinds of objects, which may in turn bear different semantic roles. In the hunter shot a duck, we have an Agent and an affected Patient, while in the hunter shot two arrows, the roles are Agent and Theme. Nevertheless, predicates and verbs are, expectedly, not the whole story. First, case (or similar) marking of arguments (flagging) may vary with one and the same



Recent developments and open questions in the field of semantic roles

verb, which has consequences for the semantic role assignment. Two illustrative examples of this are given in (4) and (5): Finnish (p.k.) (4) a. Henkilö heitt-i pallo-n yksilö-n luo. person[nom] throw-3sg.pst ball-acc individual-gen to ‘A person threw a ball to the (vicinity of) an individual.’ b. Henkilö heitt-i pallo-n yksilö-lle. person[nom] throw-3sg.pst ball-acc individual-all ‘A person threw an individual a ball.’ Lezgian (Haspelmath 1993: 292) (5) a. Zamira-di get’e xa-na. Z.-erg pot break-aor ‘Zamira broke the pot.’ b. Zamira-di-waj get’e xa-na. Z.-obl.stem-adel pot break-aor ‘Zamira broke the pot accidentally/involuntarily.’

In (4a), the denoted event involves an Agent, a Theme, and a (vicinal) Goal, the latter marked with the postposition luo ‘to’. In (4a), the only reading is that the ball has been thrown to the vicinity of the Goal, with ‘individual’ not able to bear the role of Recipient. In (4b), for its part, the third argument bears allative coding and its role has changed to Recipient instead of Goal (understood as an endpoint of transfer). In other words, the Finnish verb heittää ‘throw’ allows both Goal and Recipient roles, depending on the third argument referent. The only difference between (4a) and (4b) lies in the flagging of the third argument; in both (4a) and (4b) its referent is equally animate. In Lezgian, the Agent coding varies between ergative and adelative, with semantic consequences. With the ergative, as in (5a), we are dealing with a typical Agent (a Proto-Agent in Dowty’s terms), while in (5b), the Agent is responsible for the event, but without instigating it intentionally. In both Finnish and Lezgian, the verbs in question thus allow their arguments to take different formal manifestations with different semantic roles associated with some of their arguments. In these cases, the semantic role of an argument is a combination of verbal semantics and formal marking; the verb sets the frame, but case marking specifies the role in question. An important question related to this is whether we need formal evidence or not to distinguish between different roles. In turn, this is related to the question of whether the same roles can be distinguished for all languages, i.e., are roles universal, or should we restrict explicit distinctions to cases where we can base our claims on clear formal evidence? For example, can, or should, we distinguish between Recipients, Patients, and Addressees in primary object languages

7

8

Seppo Kittilä and Fernando Zúñiga

in which all of these receive identical formal treatment? Can we speak of distinct roles because there are numerous languages in which these roles are separated by formal cues? Or should we lump all the roles that receive identical coding together just because they are coded alike? Moreover, there are languages in which the intuitively plausible roles of Goal and Source are not explicitly distinguished, but the distinction may be based, for example, only (or at least primarily) on verbal semantics. The distinction may also be only pragmatic in that a given language lacks both grammatical and lexical means to distinguish between the roles, but the intended reading is inferable only from contextual cues (see Wälchli & Zúñiga 2006 for a more detailed discussion of this). Again, there are numerous languages in which the distinction is explicit, and we conceptualize motion events as obligatorily involving a source and a goal, but they can often be left out from linguistic expressions, as in she is running. Can we thus also postulate these roles for languages in which we have no formal evidence for our claims? Do motion verbs involve goals and sources irrespective of whether the distinction is ever made explicit? In addition, context — which is of course closely related to lexical semantics — is important with regard to what kind of semantic role an argument bears. Illustrative examples can be found in Romance and Slavic languages, as well as in some Germanic languages, where the referent of a datively coded argument may be a Beneficiary or a Maleficiary, depending on the context. In Finnish, the interpretation of the allative displays similar variation. Consider: (6)

Finnish (p.k.) Hän tek-i tämä-n minu-lle. 3sg do/make-3sg.pst this-acc 1sg-all ‘S/he did/make this for/to me.’

The allative coding of the argument in (6) implies that the participant in question is indirectly affected by the denoted event without being its primary target. Nevertheless, the exact nature of this indirect affectedness may be deemed beneficial or detrimental. The specific interpretation of such clauses follows from the lexical semantics of the arguments and the verb, as well as contextual factors (like the explicit antecedent of tämän ‘this’ in a preceding clause or an implicit one in the preceding discourse). The Finnish example in (6) may have two readings depending on the context and the reading of the verb tehdä ‘do/make’. In the first reading, ‘s/he made this for me’, the allatively coded argument refers to a (Recipient‑) Beneficiary: something concrete has been transferred to the Beneficiary’s sphere of control. The second possible reading of (6) is ‘s/he did this to me’, in which case the allative codes a Maleficiary: nothing concrete is transferred, but the Agent has done something that has detrimental consequences for the Maleficiary. This latter



Recent developments and open questions in the field of semantic roles

reading is possible, for example, if the subject referent has betrayed the referent of the allatively coded argument. More examples of this kind are discussed by Västi and Kittilä (this volume). In the examples discussed so far, verbal semantics has been at least to some extent responsible for the semantic role assignment of a given construction. However, as Västi and Kittilä (this volume) show, semantic roles can also be defined in verbless constructions. In these cases, semantic role assignment crucially depends on the other elements available. One of the arguments of Västi and Kittilä is that verbs may be omitted whenever enough information is retrievable from other cues. Typical examples are provided by constructions with the allative or ablative, both of which have semantic content and can thus function as “(quasi‑)predicates” (Siro 1964: 26–29, Västi and Kittilä this volume) in the given constructions. Consider: Finnish (p.k.) (7) a. Tutkija-lle palkinto. researcher-all prize[nom] ‘A researcher [won/got] a prize.’ (Lit. ‘to a researcher a prize’) b. Poliisi-lta myrskyvaroitus. police-abl storm.warning[nom] ‘The police [issued] a warning about a storm.’ (Lit. ‘from the police a storm warning’)

In (7), the semantic roles of both arguments are clear even in the absence of a verb. One of the arguments is explicitly coded for its role (by the allative or ablative), and the role of the zero-marked nominative argument can be inferred on the basis of that, very much in the same way that the role of the unmarked argument is inferred from the form of the marked argument in e.g. (5), where we know that, based on the ergative/adelative coding of the Agent, the unmarked argument must refer to the Patient. Moreover, it is important to note that in cases such as (7), the contribution of the verb is less important, and the case form of the explicitly coded argument is more directly responsible for the semantic role assignment. The verb, which is nevertheless often explicitly present, rather specifies the nature of the denoted act of transfer, and may thus have only minor consequences for the semantic roles of the arguments. Consider: (8)

Finnish (p.k.) Tutkija-lle annettiin/lahjoitettiin/myönnettiin palkinto. researcher-all give.pass/donate.pass/award.pass prize[nom] ‘A researcher was given/donated/awarded a prize.’

In (8), the semantic roles of the arguments remain the same despite the changes in the lexical verbs denoting the events in question; irrespective of the verb, the

9

10

Seppo Kittilä and Fernando Zúñiga

semantic roles borne by the explicitly present arguments are Recipient (tutkija ‘researcher’) and Theme (palkinto ‘prize’). This is due to the fact that the semantics associated with the allative case as well as the lexical semantics of the arguments present make only one semantic role assignment possible.5 Further examples of constructions that may lack a verb due to an obvious semantic role assignment include motion (which is semantically close to transfer) and possession (possession is in most (or all?) languages also expressed adnominally). In (8), semantic roles are determined by a semantically specific element whose contribution may even override the verb’s contribution, as shown in (8). In addition, there are also events where the semantic roles are less clear, which has evident consequences for their coding. Meteorological events are one case in point (cf. Eriksen et al. 2010). It is not an easy task to decide whether the water coming from the sky in English it is raining is best viewed as a Theme, an Agent, a Patient, or rather as something completely different. It seems that such meteorological expressions typically have semantically rather empty verbs — at least in terms of argument structure — that have mainly grammatical functions (such as the expression of tense, mood, etc.). In other words, similarly to (8), the arguments present are primarily responsible for the interpretation of a meteorological expression, especially in languages labeled as ‘argument-type’ by Eriksen et al. (2010). The omission of a verb does not affect the reading of the clause in any significant way, even though depending on the language this may affect the grammaticality of a given construction. A typical example is provided in (9): (9)

Korean (Jae Jung Song, p.c.) Pi-ka /nwun-i /wupak-i o-nta. rain-nom /snow-nom /hail-nom come-plain.ind ‘It is raining/snowing/hailing.’

The semantically more important element of (9) is clearly the noun; the verb has primarily grammatical functions. In other words, the verb is not relevant in terms of arriving at the intended reading of the given construction. In (9), the notion of semantic roles is, as noted above, trickier, but meteorological events are mentioned here in order to underline the irrelevance of verbs in certain constructions. The roles of optional obliques (so-called peripheral roles) differ from arguments in that their roles are much less directly (if at all) inferable from verbal semantics. The semantic link between verb and obliques is especially insignificant in the case of location and time expressions that are in principle common to all events regardless of their nature, because all events occur in time and space. This 5.  In this context, it may be interesting to note that ‘give’ as a lexical verb is missing in some languages, as in Amele (Roberts 1998).



Recent developments and open questions in the field of semantic roles

means that the semantic role of an oblique is determined by its own lexical semantics in these cases, which suggests that formal specification is more important for obliques than for arguments. There are languages for which this holds (e.g., Finnish), but there are also languages in which temporal and spatial expressions are zero-marked, which is rather unsurprising given that the semantic role of such expressions can be retrieved from their lexical semantics, at least in most cases, without other cues (see Creissels & Mounole 2011: 158f for the special treatment given in some languages to some nominals like geographical names and nouns referring to humans). Also in English some temporal expressions are zero-marked, while many other obliques receive overt coding, cf. as in I put the book back on the shelf [in his studio] [Ø last week].6 Moreover, verbs may have lexicalized certain peripheral roles as a part of their semantics (more or less explicitly). For example, cutting typically requires an instrument, and any argument present referring to something that can be used for cutting will be interpreted as an instrument regardless of argument coding. An example of this is provided in (10): (10)

Tukang Besi (Donohue 1999: 259) Ku-simbi-ako te tuha-su te sede te kabali. 1sg-slash-appl core family-1sg.psr core taro core machete ‘I slashed at the taro (with a machete) for my family.’

The example in (10) includes three identically coded core arguments. This could cause ambiguity, for example, if all arguments were animate, and animacy could not be used to determine which argument bears which role, but in (10) the semantic role assignment is clear. This follows from the inherent semantics of the lexical arguments and the verb. In (10), ‘family’ is the only candidate for the beneficiary role (introduced to the clause via applicativization) due to animacy (the Agent is expressed only on the verb). The verb -simbi ‘slash’ licenses three arguments, namely an Agent, a Patient, and an Instrument. Even though both non-agentive arguments are inanimate, their semantics make it clear that sede ‘taro’ refers to the Patient and kabali ‘machete’ to the Instrument. In other words, the lexical semantics of arguments, not their formal marking, is directly responsible for the semantic role assignment in (10).

6.  Notably, formally differentiated marking and zero marking with spatial and temporal expressions can be found in different phenomena of one and the same language, e.g. in is used in English I went to London in 1985 but not in I went to London (*in) last year. By a similar token, spatial at, to, and from behave differently in declarative sentences, where they are normally used (e.g. I live at the coast, I’m traveling to the coast, and I’m traveling from the coast), and in nonpolar questions with where: where do you live (*at)?, where are you traveling (to)?, and where are you traveling *(from)? We are grateful to an anonymous reviewer for pointing this out to us.

11

12

Seppo Kittilä and Fernando Zúñiga

Jackendoff (1990) distinguishes between two tiers of semantic roles that he labels thematic and action tiers respectively. For example, in the sentence Pete threw the ball the two tiers are as follows (Jackendoff 1990: 126): (11) Pete threw the ball. Source Theme Actor Patient

(thematic tier) (action tier)

The thematic tier deals with motion and location and the action tier with ActorPatient relations. Note in (12) that the thematic role Goal does not have a corresponding role on the action tier: (12) Bill entered the room. Theme Goal Actor —

(thematic tier) (action tier)

In (12), we thus only have one acting participant (Actor) that is involved in a motion event. On the thematic tier, for its part, we also have a Goal (there cannot be motion without it), but the Goal is not seen as an affected patient. This underlines the different nature of the two tiers. In this section, we have briefly discussed the ways in which semantic roles assigned to arguments may be motivated. In many cases, the verb is primarily responsible for this and the function of arguments is to specify the identity of the participants in the denoted event. Moreover, there are instances in which other elements play a central role in this, which often happens with semantically specific case forms. In yet other cases, it is primarily the lexical semantics of arguments that determines their semantic roles. It is, however, important to bear in mind that even though we have shown that there are cases in which other formal cues or the lexical semantics of arguments override the verbs in relevance, it is not our goal to downplay the relevance of verbs. Verbs are the most evident and the least ambiguous source of the arguments’ semantic roles, and they are necessary when the intended reading is not inferable from other cues. For example, cases such as (13) would be practically impossible to interpret without a verb present in the clause: Finnish (p.k.) (13) a. isä lapse-n eilen father child-acc yesterday Lit. ‘a/the father a/the child yesterday’ b. opettaja muurari-a aina teacher bricklayer-part always Lit. ‘a/the teacher a/the bricklayer always’



Recent developments and open questions in the field of semantic roles

In principle, the examples in (13) are similar to those in (7), as the given construction lacks a verb in both cases. Nevertheless, the crucial difference between (13) and (7) lies in the fact that the arguments present in (13) appear in grammatical cases (nominative, accusative, and partitive in Finnish), which makes them semantically rather vacuous, and a combination of two grammatical cases fails to provide us with enough information for the intended reading of a construction. In (7), the arguments coded by semantic cases (ablative or allative) are responsible for the semantic roles of arguments, while in (13), a verb would be needed, as the arguments are semantically rather void. The relevance of verbal semantics and formal cues are inversely proportional to each other; when the verb is semantically rich and the arguments rather vacuous, the verb is the central element, while arguments gain additional relevance when they bear semantic contents of their own. 4. Different and related roles, and role subtypes Most scholars have argued, either explicitly or implicitly, in favor of applying Occam’s razor to the question of how many semantic roles need to be distinguished in descriptive and theoretical studies. More often than not, semantically relevant differences in formal coding (e.g. non-allomorphic variation in case marking patterns) arguably justify postulating different roles; the Spanish prepositions desde ‘from’ and a ‘to’ in the motion event portrayed in (14) are a case in point. The former encodes the Source while the latter marks the Goal: (14)

Spanish (p.k.) Corrí desde el portón a la casa. run.1sg.pfv.pst from def.sg.m gate to def.sg.f house ‘I ran from the gate to the house.’

By a similar token, the comparatively rich inventories of local cases / adpositions found in Uralic languages can be used to argue in favor of distinguishing between related notions in a detailed way. In (15), for instance, luo codes the general vicinity of Pekka, while viereen is a bit more specific, meaning that the ball is thrown beside Pekka and not, for example, behind him: Finnish (p.k.) (15) a. Heit-i-n pallo-n Peka-n luo. throw-pst-1sg ball-acc P.-gen to ‘I threw the ball to / to the vicinity of Pekka.’ b. Heit-i-n pallo-n Peka-n viereen. throw-pst-1sg ball-acc P.-gen to/beside ‘I threw the ball to / beside Pekka.’

13

14

Seppo Kittilä and Fernando Zúñiga

Such a line of reasoning leads to a number of issues that are frequently glossed over or at least treated in a conspicuously atheoretical fashion. For instance, is Goal one general role with potentially more specific subroles, or are the specific relations the actual semantic roles while Goal is a general hyper-role, so to speak? A similar question that has received some attention in the literature is whether different kinds of causing participants are to be considered separate roles: are animacy, volition, and the like best regarded as possible parameters of agents, or rather as qualities or even prerequisites of prototypical agents? Do case syncretisms or subspecification phenomena in particular languages suggest that the roles in question are conceived of as more tightly interconnected than in languages that sharply and consistently distinguish their formal coding? Among the scholars that have given principled answers to such questions we find those working in the RRG tradition. As mentioned in Section 2 above, this framework distinguishes participant roles (PR, which are notional non-technical formulations for the functions “actors and props have […] in a play,” Van Valin & LaPolla 1997: 84) from semantic roles sensu stricto (SR, which are technical but predicate-specific) and thematic relations (TR, which are obtained by generalizing over individual predicates, i.e. by considering different argument positions in the “logical structure” of the clause). Thus, there is a difference in this framework between what RRG treats as the participant roles of agent (e.g. Leslie broke the glass on purpose), effector (e.g. Leslie broke the glass accidentally), and force (e.g. the flood washed away the village) — defined notionally along the parameters of purposeful instigation and animacy — and the thematic relations agent and effector (in small capitals in their formalism), which are defined as follows: (16) a. DO (x, [do′ (x, […])]) b. do′ (x, Ø)

(x = agent) (x = effector)

Crucially, force is not a separate thematic relation here; it is simply an inanimate effector. Van Valin & LaPolla explicitly say that “[a]gents are always a type of effector semantically, [which] means that agent is in effect an overlay over other, more basic thematic relations” (1997: 118). The TR instrument is not defined separately either, since it is a manipulable force, i.e. incapable of independent motion and action and “under the control of another effector” (1997: 121). This contrasts with the solution proposed by Rozwadowska (1988: 159). In her view, so-called Affected Agents (“Agents of monotransitive verbs that undergo some change, traditionally referred to as Agents and Themes at the same time,” e.g. in John rolled down the hill) are a separate thematic relation vis-à-vis Agents (“Agents of prototypical Agent-Patient verbs,” e.g. in Leslie killed the gunman) and Instruments. Næss (2007: Chapter 5) proposes an analysis that is close in spirit to Rozwadowska but distinguishes between Agents, Forces, and Instruments based



Recent developments and open questions in the field of semantic roles

on her features [±volitionality] and [±affectedness] (her third feature, [±instigation], has the same positive value for all three thematic relations). Notice that the bolt of lightning in (b) below is an effector in RRG but a subtype of Force for Næss: (17) a. Agent [+vol] [−aff] b. Force [−vol] [−aff] c. Instrument [−vol] [+aff]

John broke the window on purpose. John broke the window accidentally. The bolt of lightning broke the window. The hammer broke the window.

It is far from being a coincidence that features such as volition, affectedness, instigation (and perhaps also control, not included as a separate category here) have been chosen as decisive features by Rozwadowska and Næss in their studies. As the examples above show, these features enable us to distinguish between core roles, such as Agent, Force, and Patient. Instigation and affectedness are both highly salient features, the former being associated with the Agent role and the latter with the role of Patient; many basic events (in the Goldbergian sense: ‘salient events that cause a change in the world’, see Goldberg 1995: 39–40) include either or both of these roles. On the other hand, a theory of semantic roles built on features such as gender and number would probably fail to explain central differences between the core roles, even though these features may have relevance for the formal coding of roles. For example, in German, only singular masculine nouns appear with a distinctly accusative-marked article, but it would be rather counterintuitive to claim that the referents of masculine nouns are more typical Patients than the referents of feminine or neuter nouns. Other roles have received comparatively less attention in the general literature but figure somewhat prominently in specialized studies. Recipients are probably best viewed as future Possessors, but these two roles are usually viewed as roles in their own right. Intended Recipients are arguably a subtype of Beneficiaries, but so are Deputative/Substitutive Beneficiaries, and all of these are typically treated as subtypes of a single, somewhat generic, role. Animate and inanimate Goals are commonly treated in a parallel fashion in spite of their notable and cross-linguistically robust formal differences (see Kittilä & Ylikoski 2011 for a discussion in the context of Uralic). Motion events are usually distinguished from other manipulation situations, thereby leading to Theme being a different role from Patient, but change-of-state predicates are one single class, leading to affected Patients and effected Patients not being two separate roles (but see Hopper 1985 for cases in which these two types of Patients receive different coding). It indeed seems adequate to treat argument-related semantic features like animacy as parameters, rather than as criterial features, of particular semantic roles. By contrast, general predicate-related features like volitional instigation, control, and affectedness (and possibly others, as well as perhaps different subtypes

15

16

Seppo Kittilä and Fernando Zúñiga

of control and affectedness) are probably best seen as criterial. How to best treat predicate classes (physical manipulation, motion, communication, cognition, relation, change of state, etc.), however, does not seem to be a question with a clear answer, at least not for a theory of semantic roles that aims at being cross-linguistically applicable. 5. How much and what kind of information can we gather from formal marking patterns? We discussed some factors relevant to defining semantic roles in Section 3. In this section, we will discuss the interpretation of semantic roles from a somewhat different perspective, namely from the viewpoint of formal marking patterns. We will illustrate in more detail how formal means affect the reading of clauses. The discussed features comprise case marking (including coding by adpositions), constituent order, verbal cross-reference, and changes in verb morphology. First, case (and/or similar) marking is central to the coding of semantic roles (and thus arguments), and there are even cases where case marking is more central than verbal semantics in that semantic roles may be assigned even in the absence of a verb (see (7) and (8)). Cases have been divided into grammatical (or structural) and semantic cases, depending on their semantic specificity. Grammatical cases are semantically rather empty markers whose interpretation depends largely on the verb. Typical examples are illustrated in (18) and (19): Finnish (p.k.) (18) a. Lapsi rikko-i maljako-n. child break-3sg.pst vase-acc ‘The child broke a/the vase.’ b. Lapsi näk-i kissa-n. child see-3sg.pst cat-acc ‘The child saw a/the cat.’ Tshangla (Andvik 2010: 125) (19) a. Gopen-gi apa she-wa. chief-agt father kill-nom ‘The chief killed father.’ b. Ro-ki gari giti-rang ma-thong-ma giwala. 3-agt car when-emph neg-see-nom cop ‘He had never seen a car.’

In (18a) and (19a), we are dealing with a prototypical transitive event including an Agent and a Patient, while (18b) and (19b) comprise an Experiencer and a



Recent developments and open questions in the field of semantic roles

Stimulus. Despite the differences in semantic roles, the case marking in the given constructions remains unchanged and the case markers employed are thus semantically rather vacuous. In other words, if the accusative would code only affected targets of highly transitive events, it could not appear in (18b), where it codes a stimulus. Rather, the function of the markers in (18) and (19) is to make clear who is doing what to whom (i.e. which argument refers to the Agent/Experiencer and which to the Patient/Stimulus). In (18) and (19), formal features override semantics and we cannot infer the semantic role of an argument directly from its coding. Canonical examples of semantic cases are found in (20) and (21): Finnish (p.k.) (20) a. Liisa laitto-i kirja-n pöydä-lle / laatikko-on. Lisa put-3sg.pst book-acc table-all box-ill ‘Lisa put the book on the table/into the box.’ b. Liisa anto-i kirja-n Kalle-lle /*Kalle-en. Lisa give-3sg.pst book-acc Kalle-all /*Kalle-ill ‘Lisa gave the book to Kalle /*into Kalle.’ (21)

Yidiñ (Dixon 1994: 59) Waguja-ŋgu jugi-ø gunda-l galba:n-da. man-erg tree-abs cut-pres axe-instr ‘The man is cutting a tree (with an axe).’

The examples from Finnish illustrate the use of two locative cases, allative and illative. The relation of these cases to a specific semantic role is more direct than, for instance, that of accusative or nominative. For example, in (20a), either the allative or the illative case is used depending on whether the entity in question is placed on a flat surface or into a container. In (20b), in turn, only the allative is possible, because only the allative can describe transfer of a Theme to the vicinity of a landmark (thus including the Recipient’s sphere of control), while the illative is more concrete in its meaning. This makes the illative incompatible with the expression of a transfer event. This also underlines the semantically more specific nature of semantic cases. The example in (21), for its part, illustrates the use of the instrumental case, a very common semantic case across languages, which, expectedly, codes instruments in Yidiñ. One thing worth noting is that grammatical cases usually appear in clauses denoting highly transitive events (see e.g. Næss 2007 for a thorough discussion), while semantic cases more typically code other kinds of events and their participants. In highly transitive events, the function of the case marking of arguments is thus mainly to make clear who is doing what to whom, the roles of Agent and Patient are inherent in the verb, and case marking is not needed to underline them.

17

18

Seppo Kittilä and Fernando Zúñiga

Even though cases can be divided into grammatical and semantic according to their semantic specificity, it is important to note that the distinction is not clearcut, and a number of cases are best seen as intermediate. Some semantic cases, such as the instrumental and locatives, may appear with derived intransitives (passive, antipassive etc.) to mark a grammatically determined function, such as the Agent of the passive (see (26)). Second, some cases, especially the ergative, display massive variation in how their exact reading is determined. There are languages in which the ergative is rather clearly a grammatical case and all As bear explicit coding (see (19) above), but there are also languages in which the ergative only codes Proto-Agents (in the sense of Dowty 1991), while non-prototypical Agents receive different coding, such as a locative case. In such languages, the ergative is better seen as a semantic case due to its association with the notion of typical Agent. We should therefore not regard the ergative exclusively as a grammatical case, as there is a lot of variation in its use (see Næss 2007: 184–185 for a more detailed discussion of this). A similar claim can also be made for the accusative in languages with DOM, since the accusative does not appear with all objects. Constituent order plays a role, although a less important one than that of case marking. It often distinguishes between Agent and Patient in languages that lack morphological means of argument discrimination, but mere changes in order are not used to distinguish between, say, Agent, Instrument, and Location, as case markers are. One reason for this probably lies in the fact that it is not possible to differentiate between very many roles via word order only. For example, in prototypical transitive constructions, there are basically two slots, which means that the number of roles distinguishable by word order alone is two, while languages may have dozens of cases and adpositions. Clear tendencies are also observable in the placement of arguments or obliques; arguments referring to central roles (such as Agent and Experiencer) tend to precede constituents coding peripheral roles (such as Instrument and Location). However, changes in constituent order may be accompanied (or even caused) by other differences, such as modifications of case marking. For example, in English, the Patient-Agent order is typically attested in passives, where this is accompanied by evident changes in Agent coding. Needless to say, word order serves other functions as well, such as the expression of definiteness and/or topicality. To some extent, verbal cross-reference resembles constituent order in its relevance for semantic role coding. In many languages verbal cross-reference is limited to primary arguments (subjects/Agents) only, and all peripheral roles are lumped together in that their markers do not appear on the verb so that no distinctions between them are made. However, it should be noted that, occasionally, specialized applicativizing elements may help disambiguate roles of applied objects even though the latter are not cross-referenced on the verb, as do the Philippine(-like)



Recent developments and open questions in the field of semantic roles

voice affixes. Moreover, an argument may select its case marker from an array of distinct available options, while languages often have only one cross-reference marker available per grammatical relation, which also decreases the number of distinguishable semantic roles. Agents and other roles with an animate referent are most typically cross-referenced, but this is probably in part due to the fact that primary arguments tend to have animate referents, at least in bivalent clauses (see Bickel 2010 for a cross-linguistic discussion and Metslang 2014 for Estonian). There are, however, also cases in which verbal cross-reference can be claimed to be primarily responsible for semantic role assignment, as in (22): Choctaw (Heath 1977: 207) (22) a. Hattak-at oho:yoh(-a) Ø-Ø-pisa-h. man-subj woman(-obl) 3agt-3pat-see-pres ‘Man sees woman.’ b. Hattak-at oho:yoh(-a) Ø-i-hiyiya-h. man-subj woman(-obl) 3agt-3dat-stand-pres ‘Man waits for woman.’ c. Hattak-at oho:yoh(-a) i-Ø-nokšo:pa-h. man-subj woman(-obl) 3dat-3pat-be.afraid-pres ‘Man is afraid of the woman.’

In (22), the coding of A and O remains unchanged, while the verbal cross-reference varies according to the semantic roles of the arguments. The differences between the roles in (22) are rather minimal, and all three examples in (22) denote events that rank rather low for semantic transitivity. Despite this, the roles receive different verbal cross-reference. A further instance where verbal cross-reference may be said to play a role is found in (23): Basque (p.k.) (23) a. (Ni-k) liburu-a seme-ari eman (1sg-erg) book-def son-dat.sg.def give.partic.perf diot. aux.3sg.abs.3sg.dat.1sg.erg ‘I have given the book to my son.’ b. (Ni-k) alaba Londres-ko eskola bat-era (1sg-erg) daughter.def London-rel school one-all.sg bidali dut. send.partic.perf aux.3sg.abs.1sg.erg ‘I have sent the daughter to a school in London.’

19

20 Seppo Kittilä and Fernando Zúñiga

Basque is one of the comparatively few languages in which the verb (in (23), the auxiliary) cross-references three arguments (Agent, Patient/Theme, and Recipient). However, the third argument may appear on the verb only if it is dative-marked — a privilege reserved for animate Recipients and other arguments, but unavailable to inanimate Goals, which cannot be cross-referenced. In contrast to verbal cross-reference, markers of valency-related operations like passivization or antipassivization make an important contribution to our interpretation of semantic roles. For example, semantic cases have new functions in the derived constructions, where the intended interpretation cannot be inferred without taking verb morphology into account. Verb morphology stresses the fact that a given case form has a reading different from the expected one, which resolves possible ambiguity. Perhaps the most evident examples of verb morphology determining the semantic role assignment of arguments are provided by languages like Lardil and Warrungu: Lardil (Klokeid 1976: 552) (24) a. Mangata nethakun yaraman-in. child.nom hit horse-acc ‘The child hit the horse’ b. Yaraman neyikun (ma:nanga-n). horse[nom] hit.pass (child-acc) ‘The horse was hit by the child.’ Warrungu (Blake 1977: 25) (25) a. Bama-ŋgu gamu bidja-n. man-erg water[abs] drink-nfut ‘The man is drinking water.’ b. Bama gamu-ŋgu bidja-gali-n. man[abs] water-erg drink-antip-nfut ‘The man is drinking water.’

Case marking in Lardil follows a nominative-accusative pattern, i.e., A is coded by the nominative and O by the accusative in basic active constructions, as in (24a). In derived passives, in turn, Patient is promoted to nominative-marked subject status and the Agent bears accusative coding (24b). The mirror image of this is attested in Warrungu, where A appears in the ergative and O in the absolutive in the basic active construction, while the Patient is demoted to an ergative-marked adjunct and Agent is promoted to absolutive-marked subject status in the derived antipassive. In (24) and (25), the interpretation of the accusative or the ergative depends entirely on the derived vs. non-derived nature of the given construction.



Recent developments and open questions in the field of semantic roles

Examples in (24) and (25) illustrate the extreme instances where changes in verb morphology affect the reading of cases. Somewhat less dramatic cases are found in (26) and (27): Russian (Siewierska 1984: 162) (26) a. Devock-a my-la pol-Ø. girl-nom wash.ipfv-pst.f floor-acc ‘The girl was washing the floor.’ b. Pol-Ø my-l-sja devock-oj. floor-nom wash-ipfv-refl girl-instr ‘The floor was being washed by the girl.’ West Greenlandic (Fortescue 1984: 86) (27) a. Tuttu taku-aa. caribou see-3sg→3sg.ind ‘He saw the caribou.’ b. Tuttu-mik taku-nnip-puq. caribou-instr see-antip-3sg.ind ‘He saw a caribou.’

The Russian examples illustrate differences between active and passive, while the differences in (27) are between the basic active construction and the derived antipassive. In both languages, the instrumental case is used in the derived construction to code the demoted A or O. The function of the instrumental case is thus not the same as it is in the non-derived construction, where it codes instruments in both languages. The original semantics of the instrumental is bleached in (26– 27b), and the case can code a grammatical function. However, the rather frequent use of the instrumental (and also of locatives) in derived constructions to mark grammatical functions probably has a semantic basis. Especially the use of the instrumental in the passive is rather expected, as Agents and Instrumentals both cause events to happen, although in different ways (cf. Croft’s 2012 account of such “antecedent roles”). On the other hand, cases such as the abessive or superessive do not usually appear in derived voice constructions, since the semantic connection between the case form and the expressed function is less transparent with them. In addition to the cases discussed in (24)–(27), there are instances in which certain semantic differences become evident only in derived voice constructions. Consider: Djabugay (Patz 1991: 299) (28) a. Bama-lu gurra: du:-ny. man-erg dog.O hit-pst ‘The man hit the dog.’

21

22

Seppo Kittilä and Fernando Zúñiga

b. c. d.

Yaba-nggu djulbin guni-l. brother-erg tree.O cut-pres ‘Brother cuts a tree.’ Bama gurra:-nda du:-yi-ny. man dog-dat hit-itr-pst ‘The man hit the dog.’ Yaba djulbin-da guni-yi-ng. brother tree-loc cut-itr-pres ‘Brother cuts a tree.’

In (28a–b), the animacy of the Patient has no formal consequences, but the Patient invariably appears in the absolutive. Differences in animacy are manifested formally in the antipassive, where animate Patients appear in the dative, while inanimate Patients bear locative coding, as shown in (28c–d). In other words, the animacy of the Patient is part of the dative or the locative coding in derived constructions. We may therefore conclude that derived constructions allow us to make some differences that are not possible in basic constructions, where the form overrides meaning. A similar case is attested in German, where the passive Agent can be marked with three prepositions, von ‘of ’, durch ‘through’, and mit ‘with’, depending on the semantic nature of the Agent (see also Wälchli and Zúñiga 2006 for more examples and discussion). In this section, we have briefly discussed the interplay between form and meaning in the semantic role assignment. It is clear that in most cases neither form nor meaning alone suffices for an unambiguous interpretation of role markers, but we need to take both into account. This is especially evident in derived constructions, where case forms may have functions that are quite different from those in basic constructions. Put another way, formal features provide a hint about what the role of an argument is, but the exact role is a combination of verbal semantics, formal features, and other factors (such as context). 6. Final remarks and content of this volume Like other notions in linguistics, semantic roles both vigorously resist being abandoned and persistently defy being defined in such a way that principled theories of linguistic meaning, linguistic form, and linguistic form-function correspondences can employ them without non-trivial provisos and/or significant gaps in the range of phenomena such theories successfully cover. Intensional, extensional, and sometimes even ostensive definition attempts raise a number of interesting descriptive, comparative, and theoretical issues, as we have seen in the preceding



Recent developments and open questions in the field of semantic roles

sections, which we have not tried to address comprehensively here, let alone definitively. In the light of the ideas surveyed and the data presented above, however, it is probably unsurprising that there should be no consensus regarding the question of the exact status that semantic roles must receive in theories of language structure. The exact definitions of semantic roles vary enormously according to who has defined them and for what purpose, which naturally makes the whole notion more heterogeneous. The present paper, as well as the other articles in this volume, contribute to the ongoing discussion of the many and varied issues involved by essentially drawing the readers’ attention to some difficulties that linguistic analysis of this sort faces, as well as to novel (but not revolutionary, and certainly not radically so) ways of thinking about (or of tackling) some of the recalcitrant questions: what are semantic roles and what is their significance; how and where can they be identified; and how do they originate? First, the chapter by Creissels studies the role of the Functive across languages. This role has not been previously studied from a broad cross-linguistic perspective, and its status as a semantic role can be seen as somewhat questionable, as it cannot be given a straightforward definition using typical features such as agency and affectedness. Second, Zúñiga’s paper offers a new perspective on the rather widely discussed role of Beneficiary. Zúñiga claims that surrogation and substitution should be seen as distinct from benefaction, even though these notions are intimately related. Fedriani & Prandi examine the dative role complex in Latin and its functionally equivalent complex in modern Romance languages, thus offering a diachronic perspective to the study of semantic roles. Västi & Kittilä’s study focusses on verbless constructions, where semantic roles can, however, be identified. This chapter suggests that semantic roles are best seen as constructional properties in the Goldbergian sense, but it goes one step further in discussing constructions that are genuinely verbless — thereby naturally challenging the central role of the verb in defining semantic roles. Haspelmath et al.’s study approaches semantic roles from a novel perspective as well, starting with language-internal generalizations and only then moving to the cross-linguistic level. This chapter also shows that languages can be compared at the level of individual verb meanings, using a variant of the semantic-map method applied to micro-roles. Finally, Bickel et al. study whether and to what extent there is cross-linguistic evidence for postulating clusters of predicate-specific semantic roles such as Experiencer, Cognizer, Possessor, etc. Applying fuzzy cluster and NeighborNet algorithms to non-default case-marking patterns produces cross-linguistic evidence for role clusters around Experiencers, undergoers of body processes, and Cognizers/Perceivers, as well as around Sources and transmitted speech, but no significant support is found for any other role clusters. This chapter thus shows that there are clear differences between what had traditionally been assumed to be semantic roles to be treated on a par.

23

24

Seppo Kittilä and Fernando Zúñiga

Abbreviations abl abs acc adel agt all antip aor appl aux cop core emph erg dat

Ablative case Absolutive case Accusative case Adelative case Agent Allative case Antipassive Aorist Applicative Auxiliary Copula Core argument Emphatic Ergative case Dative

def f gen ill ind instr ipfv itr loc m neg nfut nom o obl

Definite Feminine Genitive case Illative case Indicative mood Instrumental case Imperfective aspect Intransitive Locative case Masculine Negation Non-future tense Nominative case Object-like argument Oblique

obl.stem part partic.perf pass pat pfv p.k. plain.ind pres psr pst refl rel sg subj

Oblique stem Partitive case Perfective participle Passive Patient Perfective Personal knowledge Plain indicative Present tense Possessor Past tense Reflexive Relative marker Singular Subject

References Andvik, Erik. 2010. A grammar of Tshangla. Leiden: Brill.  doi: 10.1163/ej.9789004178274.i-490 Baker, Mark. 1988. Theta theory and the syntax of applicatives in Chichewa. Natural Language and Linguistic Theory 6(3). 353–389.  doi: 10.1007/BF00133903 Bickel, Balthasar. 2010. Grammatical relations typology. In Jae-Jung Song (ed.), The Oxford handbook of language typology, 399–444. Oxford: Oxford University Press. Blake, Barry J. 1977. Case marking in Australian languages. Canberra: Australian Institute of Aboriginal Studies. Bresnan, Joan. 2001. Lexical-functional syntax. Oxford: Blackwell. Carnie, Andrew. 2002. Syntax: A generative introduction. Oxford: Blackwell. Creissels, Denis & Céline Mounole. 2011. Animacy and spatial cases: Typological tendencies, and the case of Basque. In Seppo Kittilä, Katja Västi and Jussi Ylikoski (eds.), Case, animacy and semantic roles, 157–182. Amsterdam: Benjamins. Croft, William. 2012. Verbs. Aspect and causal structure. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199248582.001.0001

Culicover, Peter & Ray Jackendoff. 2005. Simpler syntax. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199271092.001.0001

Dixon, R. M. W. 1994. Ergativity. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9780511611896

Dixon, R.M.W. 2010. Basic Linguistic Theory. Volume I: Methodology. Oxford: Oxford University Press. Donohue, Mark. 1999. A grammar of Tukang Besi. Berlin: Mouton de Gruyter.  doi: 10.1515/9783110805543

Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67(3). 547– 619.  doi: 10.1353/lan.1991.0021



Recent developments and open questions in the field of semantic roles

Eriksen, Pål, Seppo Kittilä & Leena Kolehmainen. 2010. Linguistics of weather: Cross-linguistic patterns of meteorological expressions. Studies in Language 34(3). 565–601.  doi: 10.1075/sl.34.3.03eri

Fillmore, Charles. 1968. The case for case. In Emmon Bach & Robert Harms (eds.), Universals in Linguistic Theory, 1–88. New York: Holt, Rinehart, and Winston. Fortescue, Michael. 1984. West Greenlandic. London: Croom Helm. Fried, Mirjam & Jan-Ola Östman. 2004. Construction grammar: A thumbnail sketch. In Mirjam Fried & Jan-Ola Östman (eds.), Construction grammar in a cross-language perspective, 87– 120. Amsterdam: John Benjamins.  doi: 10.1075/cal.2.03fri Goldberg, Adele. 1995. Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Goldberg, Adele. 2006. Constructions at work: The nature of generalization in language. Oxford: Oxford University Press. Gruber, Jeffrey. 1965. Studies in lexical relations. Cambridge, MA: MIT Press. Hale, Kenneth & Samuel Keyser. 1993. On argument structure and the lexicon expression of syntactic relations. In Kenneth Hale & Samuel Keyser (eds.), The view from building 20: Essays in linguistics in honor of Sylvain Bromberger, 53–119. Cambridge, MA: MIT Press. Hale, Kenneth & Samuel Keyser. 2001. Prolegomenon to a theory of argument structure. Cambridge, MA: MIT Press. Haspelmath, Martin. 1993. A grammar of Lezgian. Berlin: Mouton de Gruyter.  doi: 10.1515/9783110884210

Heath, Jeffrey. 1977. Choctaw cases. Proceedings of Berkeley Linguistic Society 3. 204–213. Hopper, Paul J. 1985. Causes and affects. In William H. Elifort, Paul D. Kroeber & Karen L. Peterson (eds.), Papers from the parasession on causatives and agentivity at the Twenty-first Regional Meeting of CLS, 67–88. Chicago. Jackendoff, Ray. 1983. Semantics and cognition. Cambridge, MA: MIT Press. Jackendoff, Ray. 1990. Semantic structures. Cambridge, MA: MIT Press. Kittilä, Seppo & Jussi Ylikoski. 2011. Remarks on the coding of direction, recipient and vicinal direction in European Uralic. In Seppo Kittilä, Katja Västi & Jussi Ylikoski (eds.), Case, animacy and semantic roles, 29–64. Amsterdam: John Benjamins. Klokeid, T.J. 1976. Lardil. In R.M.W. Dixon (ed.), Grammatical categories in Australian languages, 550–584. New Jersey: Humanities Press. Metslang, Helena. 2014. Partitive noun phrases in the Estonian core argument system. In Tuomas Huumo & Silvia Luraghi (eds.), Partitive cases and related categories, 177–256. Berlin: Mouton de Gruyter. Næss, Åshild. 2007. Prototypical transitivity. Amsterdam: Benjamins.  doi: 10.1075/tsl.72 Newmeyer, Frederick. 2010. On comparative concepts and descriptive categories: A reply to Haspelmath. Language 86(3). 688–695.  doi: 10.1353/lan.2010.0000 Patz, Elizabeth. 1991. Djabugay. In R.M.W. Dixon & Barry J. Blake (eds.), The handbook of Australian languages. Volume 4. The Aboriginal language of Melbourne and other grammatical sketches, 244–347. Oxford: Oxford University Press. Perlmutter, David & Paul Postal. 1984. Studies in Relational Grammar 2. Chicago: University of Chicago Press. Reinhart, Tanya. 2002. The theta system: An overview. Theoretical Linguistics 28(3). 229–290. Roberts, John R. 1998. GIVE in Amele. In John Newman (ed.), The linguistics of giving, 1–34. Amsterdam: Benjamins.  doi: 10.1075/tsl.36.02rob

25

26 Seppo Kittilä and Fernando Zúñiga Rozwadowska, Bożena. 1988. Thematic restrictions on derived nominals. In Wendy Wilkins (ed.), Syntax and semantics 21: Thematic relations, 147–165. New York: Academic Press. Rozwadowska, Bożena. 1989. Are thematic relations discrete? In Roberta Corrigan, Fred Eckman & Michael Noonan (eds.), Linguistic categorization, 115–130. Amsterdam: John Benjamins.  doi: 10.1075/cilt.61.09roz Siewierska, Anna. 1984. The passive: A comparative linguistic analysis. London: Croom Helm. Siro, Paavo. 1964. Suomen kielen lauseoppi. Helsinki: Tietosanakirja. Van Valin, Robert & Randy LaPolla. 1997. Syntax: Structure, meaning, and function. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9781139166799 Wälchli, Bernhard & Fernando Zúñiga. 2006. Source-Goal (in)difference and the typology of motion events in the clause. Sprachtypologie und Universalienforschung 59. 284–303.

Identifying semantic role clusters and alignment types via microrole coexpression tendencies Iren Hartmann,1 Martin Haspelmath1 and Michael Cysouw2

1Max

Planck Institute for the Science of Human History / 2PhilippsUniversität Marburg

In this chapter, we illustrate a method for identifying clusters of semantic roles by cross-linguistic comparison. On the basis of data from 25 languages drawn from the ValPaL (Valency Patterns Leipzig) database, we show how one can visualize coexpression tendencies using quantitative methods (in particular, multidimensional scaling). Traditionally, the coexpression of semantic microroles (such as the breaker and the broken thing of the ‘break’ verb, the helper and the helpee of the ‘help’ verb, etc.) has been studied for particular languages, with generalized macroroles such as “agent”, “actor”, and “undergoer” being compared across languages in a next step. We set up a conceptual space of 87 microroles based on their coexpression tendencies, i.e. the extent to which they are expressed identically (via flagging and indexing) across our languages. The individual coding means (cases, adpositions, index-sets) can then be mapped onto this conceptual space, revealing broader alignment patterns. Keywords: microrole, coexpression, semantic map, alignment type, valency

1. Introduction Semantic role notions are widely used by linguists to capture generalizations of argument coding and argument behavior across verbs within a single language. Thus, in German the distribution of Nominative and Accusative arguments in twoplace verbs like zerbrechen ‘break’, schlagen ‘beat’, and essen ‘eat’ is not random, but follows a semantic regularity: The agent (breaker, beater, eater) is coded by the Nominative case, and the patient (broken, beaten, eaten) argument is coded by the Accusative case. However, a precise semantic characterization of roles like “agent” and “patient” is not easy, and in practice linguists have often followed a

doi 10.1075/bct.88.02har 2016 © John Benjamins Publishing Company

28

Iren Hartmann, Martin Haspelmath and Michael Cysouw

semasiological (form-to-function) approach, so that they have (often unwittingly) ended up describing patients as whatever Accusative arguments share semantically. But it turns out that cross-linguistic comparison is difficult when using such semasiologically defined roles. Thus, in German one would exclude the non-agent arguments of helfen ‘help’, folgen ‘follow’, and dienen ‘serve’ from the “patient” role because they are expressed as Dative arguments, while in English the corresponding roles are naturally included. German patients and English patients are therefore no longer directly comparable, and these roles cannot be used readily to compare languages. This situation has given rise to the well-known problems in applying semantic roles (especially role hierarchies) across languages (e.g. Newmeyer 2005: 215–220). In this chapter, we adopt a different approach to cross-linguistic comparison. Instead of starting with language-internal generalizations about roles and then moving to the cross-linguistic level, we show that one can usefully compare languages at the level of the roles of individual verbs (microroles). Semantic role clusters can then be identified by studying cross-linguistic coexpression tendencies, i.e. the ways in which the individual microroles cluster with respect to their coding across a range of diverse languages. Like the more traditional approach sketched in the first paragraph, our method assumes that the coexpression of roles is not random, i.e. that it is not an accident that the breaker, the beater, and the eater are coded alike (= coexpressed) in German and English, and that the helpee, the followee, and the servee are coexpressed in German (by Dative case). Accidental homonymy may exist, of course, but repeated coexpression of the same notional elements across many different languages must indicate similarity of meaning (e.g. Haiman 1974, Wälchli 2009, Wälchli & Cysouw 2012). Our approach is thus a variant of the semantic-map method, applied to microroles of individual verbs and their coding by argument flagging (= case or adpositional marking) or argument indexing (= cross-referencing and “person agreement”). This study does not take word order as a means of coding arguments into consideration, as word order is far more difficult to compare across languages and will have to be left for future research. 2. Three levels of semantic role granularity There are at least three different ways in which role terminology can be used (and has been used in the literature). These different approaches define roles at three different levels of granularity. First, roles can be formulated at a verb-specific level. For example, verbs like ‘hit’, ‘hug’, and ‘fear’ involve different verb-specific role

Identifying semantic role clusters and alignment types via microrole coexpression tendencies

pairs, namely ‘hitter/hittee’, ‘hugger/huggee’, and ‘fearer/fearee’, respectively.1 We will refer to such roles as microroles. Secondly, one can define roles such as ‘agent’, ‘patient’, or ‘experiencer’ at an intermediate level — more abstract than microroles, but still close to the semantic content. For example, the ‘agent’ might subsume the microroles ‘hitter’ and ‘breaker’, the patient might subsume such roles like ‘breakee’, ‘hittee’, and ‘huggee’, and the ‘experiencer’ might be represented by the ‘fearer’ in our example. We will refer to the roles at this intermediate level as mesoroles. Thirdly, roles can be defined at a level that is close to the level of argument realization itself (especially by means of flagging or indexing). From this perspective a role like ‘Actor’ will subsume the agent and the experiencer from the intermediate level and the ‘Undergoer’ will subsume the patient and other roles, as for example ‘stimulus’.2 Roles at this higher level of abstraction have been called macroroles (Van Valin 2005) (or hyperroles, Kibrik 1997). The way in which microroles can be collapsed into mesoroles, and these in turn can be collapsed into macroroles is shown nicely by Van Valin (2005: 54); see Figure 1 as an illustration.3 Verbal meaning hit break hug fear

MICROROLES

MESOROLES

MACROROLES

hitter hittee

agent

breaker breakee

Actor patient

hugger huggee

experiencer

fearer fearee

stimulus

Undergoer

Figure 1.  Hierarchy of role concepts (adapted from Van Valin 2005: 54)

1.  The role names — like ‘hugger’ and ‘huggee’ — make liberal use of English word-formation rules and thus will not necessarily sound idiomatic. We use them to avoid clumsy expressions like ‘the participant that hugs’, ‘the participant that is hugged’. 2.  Lehmann (2006) has introduced a third macrorole, the indirectus, but this has not yet become widely known. 3.  Cysouw (2014) introduces yet another level of granularity of roles, viz. contextual role, which can be added even further to the left in the hierarchy or role concepts from Figure 1. The motivation for this is that even verb meanings are sometimes difficult to compare across languages, and it might be favorable in some situations to look at the coding of a role in a specific context, independently of the actual verb that is being used.

29

30

Iren Hartmann, Martin Haspelmath and Michael Cysouw

Each of these three levels has its problems and limitations. The limitation of the verb-specific level of microroles is that it has no language-internal generality at all. At the other end of the spectrum, the problem with the argument-realization level of macroroles is that it has (too) little cross-linguistic generality. As a result, linguists have typically worked at the level of mesoroles. This level appears to allow for generalizations within a language (allowing one, for example, to account for the similar behavior of ‘break’, ‘hit’, and so forth), but it also allows for crosslinguistic comparison. For example, it becomes possible to express many salient differences between languages, such as the contrast between experiencer-subject constructions (e.g. I like it) and experiencer-object constructions (e.g. it pleases me). Especially for textbook accounts, mesoroles work fairly well, and they have become the best-known type of roles. However, as we saw in the previous section, mesoroles are still strongly language-specific, and there are many roles that do not fit well into the established categories, e.g. the roles of the boldfaced arguments in (1). (1)

a. b. c. d. e. f.

The women searched for the man. (searched-for thing) The women searched the woods for the missing child. (searched place) The boy looked at the girl. (looked-at entity) The old man coughed. (cougher) My arm hurts (me). (pain locus) This man is a hunter. (hunter)

Verbs such as these may be in the minority, but they show that mesoroles, though commonly used, are not sufficient, whether for cross-linguistic comparison or for language-particular description. Moreover, it appears that those verbs and verbspecific roles that are the most problematic for description are also the ones that differ the most across languages, i.e. while these problems are precisely those cases in which comparison would be the most important, they have so far been intractable. In this chapter, we only use microroles to compare languages.4 While we are fully aware that our method does not make a particular contribution to the question of how best to describe individual languages, some of the problems in earlier 4.  One reviewer asked whether our approach presupposes that argument coding is only sensitive to the role meaning of the arguments. The answer is that our comparison only captures meaning-based similarities. Argument coding may be sensitive to other factors, such as the grammatical composition of a verb (thus, causative verbs are more likely to have causee and patient coded alike than monomorphemic verbs, cf. Malchukov 2013). Such similarities are not taken into account here, and are a possible confounding factor, though of course the great majority of our verbs are monomorphemic. Argument coding is indeed also often influenced by the referential properties of the argument (e.g. definiteness), but such properties are ignored in our coding sets (cf. Haspelmath & Hartmann 2014).

Identifying semantic role clusters and alignment types via microrole coexpression tendencies a. accusative

A

ergative

S

S

P

A

b. English helper helpee freezing person hitter hittee

P

ergative helper

helpee freezing person hitter hittee

Figure 2.  Examples of macrorole (a.) and microrole (b.) argument alignments

work have stemmed from the misguided attempt to use the same notions for description and for comparison. The microroles that we use here are thus intended exclusively as comparative concepts (Haspelmath 2010). Our approach allows for comparison on the basis of these very basic comparative concepts and does not rely on language-internal generalizations beyond the individual verb and its argument coding. 3. From macrorole alignment to microrole coexpression The coexpression5 of different roles by argument-coding elements (case markers, adpositions, person indexes) has been one of the central concerns of typology since the 1970s. This goes by the name of “argument alignment” and has normally worked with prototype-based macroroles such as A, S, P, T, and R (Dixon 1979, Comrie 1981, Haspelmath 2011). The two major types of monotransitive alignment are the accusative type, where intransitive S is coded like A (= coexpressed with A) but differently from P, and the ergative type, where intransitive S is coded like P (= coexpressed with P) but differently from A. This is generally represented in a semantic-map-like diagram as shown in Figure 2a. This analysis works well for prototypically transitive clauses that have an A and a P argument, but the approach has nothing to say with regard to other kinds of two-argument verbs. Likewise, the variation within single-argument verbs (agentive-patientive and similar systems, 5.  The term coexpression is used here for the relation between a linguistic form in a given language and several functions that are expressed by different forms in some other language. It seems that this is a terminological innovation of the present paper, but we regard it as a very natural one. It was inspired by François’s (2008) term colexicalization (for lexical coexpression). The term is convenient in a cross-linguistic context because, unlike terms such as multifunctionality or polysemy, it is completely neutral as to the implications for the language-specific description.

31

32

Iren Hartmann, Martin Haspelmath and Michael Cysouw

cf. Donohue & Wichmann 2008) is sometimes accommodated by distinguishing between Sa and Sp, but these two “sub-macroroles” are not particularly homogeneous across languages. In order to capture a broader range of cross-linguistic differences in argument coding, we need to extend the alignment or coexpression approach to a larger set of roles that can be clearly identified cross-linguistically. This is what we set out to do in the present paper: We study the alignment of the coding of 181 different microroles, which belong to 87 different verb meanings (intended to be representative of the entire verbal lexicon). For example, consider the three aforementioned verb meanings ‘help’, ‘hit’, and ‘freeze’, which behave differently in English and German: (2)

English a. SheNOM helps meACC. b. SheNOM hits meACC. c. SheNOM is freezing.

(3)

German a. SieNOM hilft mirDAT. ‘She helps me.’ b. SieNOM schlägt michACC. ‘She hits me.’ c. IhrDAT ist kalt. ‘She is freezing.’

The coexpression of the five roles is rather different in English and German, as can be seen in Figure 2b. Some further cross-linguistic differences of this sort can be seen in Table 1, which compares the coding sets6 of seven microroles in three other languages. We see that only the ‘hitter’ and the ‘helper’ are coexpressed in all three languages (see the grey shading). The ‘hittee’, ‘likee’, and the ‘helpee’ are coexpressed in two languages (Hoocąk and Chintang), and the ‘liker’ and the ‘helpee’ are only coexpressed in one language (Icelandic). Comparing such pairs of microroles and their coexpression or non-coexpression will be the basis for our study. It should be noted that the language-specific forms and their labels play no role in our comparisons, only identity of coding between different microroles. If we renamed the Icelandic Nominative and called it “Subjective”, and if we did the same with the Chintang Ergative, the two languages would not become more similar. All that counts is the language-internal identity of coding of the microroles. Our approach thus exemplifies the shift of typology from macro-alignment (using A, S, P etc.) to micro-alignment, i.e. to investigate how individual verbs are treated compared to other individual verbs. This is in line with what Bickel 6.  By coding set we refer to the sum of coding element(s) that mark an argument, i.e. a flag (case/adposition) on the NP and/or an index-set (such as subject agreement, object indexing, etc.) on the verb.

Identifying semantic role clusters and alignment types via microrole coexpression tendencies

Table 1.  Seven microroles with associated coding sets in three different languagesa microrole

Icelandic

Hoocąk

Chintang

hitter

NP-nom & sbj.V

act.V

NP-erg & V.sbj

hittee

NP-acc

und.V

NP-abs & V.obj

liker

NP-dat

act.V

NP-erg & V.sbj

likee

NP-nom & sbj.V

und.V

NP-abs & V.obj

throw goal

inn um+NP-acc

NP+eeja

NP-abs & V.obj

helper

NP-nom & sbj.V

act.V

NP-erg & V.sbj

helpee

NP-dat

und.V

NP-abs & V.obj

a 

The abbreviations used in Table 1 stand for the following: abs = Abolutive; acc = Accusative; act = Actor indexing; dat = Dative; erg = Ergative; obj = Object indexing; sbj = subject indexing; und = undergoer indexing. Words in italics are language-specific adpositions.

suggests as the characteristic feature of 21st century typology: “Linguistic diversity is captured by large sets of fine-grained variables, not by grand type notions” (Bickel 2007: 245). Our paper thus continues a line of research that was begun by Nichols (2008) for single-argument verbs, but we apply it to a broad cross-section of verbal meanings. 4. Microrole coexpression in 25 languages Our data consists of coding elements (flags and index-sets) used in a comparable sample of verbs corresponding to 87 verb meanings across 25 genealogically diverse languages from different parts of the world. The data was contributed by language experts participating in the Leipzig Valency Classes Project,7 and was collected via an electronic database questionnaire, a special database created with 7.  The languages investigated and their respective data contributors are the following: Ainu (Anna Bugaeva), Balinese (Masayoshi Shibatani & Ketut Artawa), Central Alaskan Yupik (Osahito Miyaoka), Modern Standard Arabic (Csilla Kász), Bezhta (Zaira Khalilova & Bernard Comrie), Bora (Frank Seifart), Chintang (Robert Schikowski & Balthasar Bickel), Even (Andrej Malchukov), Hoocąk (Iren Hartmann), Icelandic (Jóhanna Barðdal), Italian (Michela Cennamo), Jakarta Indonesian (Thomas Conners & David Gil), Jaminjung (Eva Schultze-Berndt), Ket (Elena Krjukova & Edward Vajda), Mandarin (Guohua Zhang & Bingfu Lu), Mandinka (Denis Creissels), Mapudungun (Fernando Zúñiga), Nen (Nicholas Evans), N|uu (Martina Ernszt, Alena Witzlack-Makarevich & Tom Güldemann), Sliammon (Honore Watanabe), Sri Lanka Malay (Sebastian Nordhoff), Xârâcùù (Claire Moyse-Faurie), Yaqui (Zarina Estrada, Mercedes Tubino, and Jesús Francisco Villalpando), Yoruba (Joseph Atoyebi), Zenzontepec Chatino (Eric Campbell). We are very much indebted to all our data contributors. Without their expertise and data this study would not have been possible.

33

34

Iren Hartmann, Martin Haspelmath and Michael Cysouw

FileMaker Pro™. The data is directly accessible in a structured way. It is also comparable, as the verbs were elicited in response to a list of 87 pre-defined verbal meanings, and also because the valency information was recorded in a standardized way (to access the data, please go to www.valpal.info (Hartmann et al. 2013)). Our 87 verb meanings can be found in Appendix 1 in a manner similar to how they were presented to the data contributors. They are presented in three columns: (i) the meaning label, (ii) the role frame, and (iii) a “typical context”.8 The data was exported from FileMaker Pro™ to be used in the statistical environment R (R Development Core Team 2012).9 The 87 different verb meanings include in total 189 different microroles, but because of missing data (i.e. roles for which we do not have information from more than 5 languages) we only use 181 microroles for the current paper. Because the verb meanings correspond closely across languages, the microroles of the verbs can be seen as equivalent across languages and we can thus compare them in terms of their overt coding properties (flagging and indexing). As noted in the preceding section, despite the diversity of the actual coding elements, we can still ask which of the microroles are coded in the same way (i.e. coexpressed) within each single language. Basically, the number of coexpressions between two microroles, averaged over the 25 languages, provides an estimate of the similarity between the microroles. This approach yields an objective metric on microroles, motivated by the overt form of the argument coding. Finally, we can then produce a visualization of this metric, representing a semantic map of microroles across languages (cf. Cysouw 2010). In this chapter, the metric on microroles was defined as follows. First, when the coding set of a microrole consists of two coding elements (a flag and an index), we separated each coding set into its elements. For example, in Icelandic, the ‘hitter’ microrole is coded by Nominative case (a flag) and Subject agreement on the verb (an index-set). The similarity between two microroles within a language was then defined as the number of coding elements (not sets) shared by the two, divided by the maximum number of elements in either of the two. Second, for each pair of microroles, the average similarity was taken from all languages for which data was available. Third, we computed the correlation between each of the language-specific metrics and the average metric over all languages (simply using a Pearson correlation between the similarities). The higher this correlation, 8.  Since the English verbs used as meaning labels sometimes have different meanings, we have added a sentence for each verbal meaning that makes the intended meaning clear. These sentences are not crucial, they are just intended to help the contributors find a context for their counterpart. 9.  We would like to express our sincere gratitude to Hans-Jörg Bibiko for all of his help in this area.

Identifying semantic role clusters and alignment types via microrole coexpression tendencies

Figure 3.  First two dimensions of a multidimensional scaling of the weighted average metric on the 181 microroles, showing clear semantic separation into meso- or macroroles

the more a specific language looks like an ‘average’ language (i.e. ‘average’ for this study). To be able to better distinguish diversity, we then computed a weighted average metric by weighting each language using the negative logarithm of the correlation coefficient (this being a very rough approximation of the amount of information of the average inherent to each language separately). 5. Clustering roles As a visualization of this weighted average metric on the microroles, we used a classic ‘metric’ multidimensional scaling (MDS). The first two dimensions of the MDS are shown here in Figure 3, which represents a kind of semantic map. Not all 181 microroles are readable because of overlap, but some major role clusters are still discernible. Basically, what happens is that the average coexpression across many languages identifies clusters of microroles which approximately represent the traditional notion of mesoroles (or even macroroles depending on how large one considers a cluster to be) (see Cysouw 2014 for a detailed discussion). To the left in Figure 3 there is a cluster of agent-like roles, at the bottom there is a

35

36

Iren Hartmann, Martin Haspelmath and Michael Cysouw

zero jiʔį`+NP lóʔó+NP

Figure 4.  Distribution of the three coding elements in Zenzontepec Chatino

cluster of patient-like roles, and at the top right there is a cluster of instruments. Recipients and goals are found in between patients and instruments. Figure 3 can be regarded as a more detailed version of Figure 2b above, except that the spatial closeness of the microrole labels in Figure 3 was computed on the basis of the coding sets of the 181 microroles, while the arrangement of the five microroles in Figure 2b was done manually. Thus, while the intermediate position of the ‘freezing person’ in Figure 2b serves to make the figures easy to read, the fact that the ‘fear stimulus’ occurs right in the middle of Figure 3 reflects the empirical fact that this role is coded about equally as often (i.e. flagged and/or indexed) as the ‘peeler’ and ‘wiper’ (to the upper left) and as the ‘tellee’ (to the right) and the ‘hidden thing’ (at the bottom). To the extent that cross-linguistic coexpression tendencies are due to semantic similarity, we can say that Figure 3 gives us an approximate, but objective representation of the language-independent semantic similarities between the microroles. No subjective judgment of semantic similarity (e.g. between the ‘hidden thing’ and the ‘killee’) in any particular language is involved.

Identifying semantic role clusters and alignment types via microrole coexpression tendencies Balinese

Bora

zero NP-acc NP-adl NP-abl NP-instr

zero aji+NP ke+NP uli+NP

Hoocąk

Jaminjung

act.V und.V -eeja/LOC unmarkedNP

NP-ergsubj.V NP-abs obj.V NP-abs subj.V NP-datV=pro.dat NP-erg/instr NP-abs NP-loc/all NP-all

Figure 5.  Four additional languages showing different distributional ranges of coding sets

6. Mapping languages on the microrole map In a next step, we can use the two-dimensional layout from Figure 3 as a base map for the comparison of the role marking of individual languages. As a first example of this method, consider the structure of Zenzontepec Chatino role marking in Figure 4. To produce this graphical display, we used the following method. First, the location of the 181 microroles was taken from the multidimensional scaling as shown in Figure 3 (displayed in Figure 4 as small light grey circles). The available data from Zenzontepec Chatino is then plotted on top of this, using different symbols for different language-specific coding elements. In this language, there are only three different argument coding elements: zero (mostly for agent-like arguments) and two flags (prepositions), jiʔį̀/j- (for a large variety of semantically disparate arguments, glossed prep) and lóʔò (for instrument-like arguments), as illustrated in (4).

37

38

Iren Hartmann, Martin Haspelmath and Michael Cysouw

(4)

Zenzontepec Chatino (Campbell 2013) Nkajnyà chu kíʔyùú j-nà niʔií lóʔò lítúu. cplv.build nom.hum male.dem prep-art house with adobe ‘The man built the house with adobe.’

In Figure 4, these three coding elements are identified by squares, circles, and triangles, respectively. The distribution of these three symbols clearly shows clusters, with the squares being used for the agent-like microroles on the left-hand side, the circles for patient-like microroles as well as goal-like, addressee-like, and recipient-like microroles, and the triangle being used for instrument-like microroles in the upper right corner. To make it easier to see this distribution, we used an R package called Krig to draw lines around the areas of highest density for each coding element. These lines are reminiscent of the boundary lines traditionally drawn in semantic maps (cf. Haspelmath 2003) as well as the contour lines on a traditional geographical map. However, technically speaking, the lines in our figures have a different rationale. Instead of drawing boundary lines precisely around all points that are coded by a particular coding element, the lines here represent probabilistic indications of the regions in which particular elements predominate. This also explains why there are various coding elements to be found within the ‘wrong’ lines, for example, some triangles in the area of patient-like coding. To be precise, the lines represent three different probability distributions (one for each construction) in two-dimensional space, indicating which parts of the figure are more likely to be coded by each element. To show these three probability distributions in one figure we have only drawn lines indicating the probability of 35% (with two thinner lines indicating 32% and 29%, just to visually indicate the gradient nature of these lines). For all lines to be comparable, these probabilities are kept constant throughout all figures in this chapter. To infer the probabilities we made use of kriging, a geostatistical method to interpolate distributions in space.10 In our case, we interpreted the points of Figure 3 as points in space. Then, each point was given a height of one when a specific coding element was present, and a height of zero when a different coding element was attested. Missing data for individual roles was ignored (this can be seen in the grey circles of the base map that are not accompanied by a black circle, triangle, or square). This distribution of high (one) and low (zero) points was then interpolated as ‘hills’ in space, and the lines were drawn at a height of 0.35, 0.32, and 0.29. In Appendix 2 all of the different distributions of the coding devices from all 25 languages are shown. Because 10.  We used the Krig function for the library fields (Furrer et al. 2012) as provided for the statistical environment R (R Development Core Team 2012).

No agent/patient

Nen Jaminjung Alaskan.Yupik Chintang Hoocąk Sri.Lanka.Malay Icelandic Ket Zenzontepec.Chatino Even Italian Bora Mapudungun Ainu Sliammon Arabic Yaqui

Jakarta.Indonesian Mandarin Nluu Balinese Xârâcùù Mandinka Yoruba

Bezhta

Identifying semantic role clusters and alignment types via microrole coexpression tendencies

Tripartite

Agent/patient

Figure 6.  Hierarchical clustering of similarities in microrole coexpression (i.e. alignment similarities)

the plots were drawn completely automatically, using exactly the same settings for each plot, the plots are directly comparable across languages. In Figure 5, we see the distributions of the coding sets in four additional languages, with distribution lines to help us recognize the clusters of coding sets. We see that Balinese is a language that lacks a difference between overt agent and patient coding, while Bora makes a clear agent-patient distinction (Nominative vs Accusative case marking).11 In Hoocąk, the patient domain is bigger, reflecting its active-stative (Split-S) alignment. Finally, Jaminjung is a double-marking language with a tripartite pattern, distinguishing between transitive agents (Ergative case and Subject indexing), intransitive subjects (Absolutive case and Subject indexing), and transitive patients (Absolutive case and Object indexing). (This kind of tripartite alignment is not generally recognized in the typological literature, because flagging and indexing are considered separately. However, in our approach the full coding set of each microrole is taken into account, both the flag and the index.)

11.  Of course, this does not necessarily mean that agents and patients are more difficult to distinguish by Balinese hearers, because word order is an additional coding device that we did not take into account. Thus, we are not necessarily claiming that agents and patients are semantically more similar in Balinese than in Bora. All we are claiming is that they are treated more similarly in terms of flagging and indexing.

39

40 Iren Hartmann, Martin Haspelmath and Michael Cysouw

7. Clustering languages: Alignment types The visualizations shown in Figures 4 and 5 not only present a method for the visual inspection of the structure of different languages, we can also use it to compare languages quantitatively. Instead of attempting to classify the 25 languages into discrete types, we quantify the pairwise similarity between all languages. In this way we are able to investigate global similarities between languages (i.e. to develop a language typology), without ignoring the interesting and important differences between the languages. Instead of discretely classifying languages into types (i.e. making a ‘partition’), we provide a metric for languages, specifying precisely how similar each language is to each and every other language. We can then use the metric to provide a partition, or any other simplification of the actual similarities (e.g. a hierarchical clustering as shown in Figure 6 below). The actual comparison between two languages was performed as follows. Basically, we made language-specific metrics on all 181 microroles, using just the coding of each single language. For each pair of microroles, only the coding similarity from a single language was used. Therefore, when two microroles used the same coding, their difference was zero, while in the case of two completely different coding devices, the difference was set at one. This was performed for all 181*180/2 = 16290 pairs of microroles, and for all 25 languages. We then used a weighted Pearson correlation to establish the similarity between two languagespecific lists of these 16290 pairs of microroles. We used the weighting to control for the fact that the microroles are not equally distributed across all possible functions. As can immediately be discerned from Figure 3, there are many more agent-like and patient-like microroles than instrument-like or goal-like roles. Any unweighted comparison would thus strongly favor whatever differences exist in these highly frequent kinds of roles. To counteract this overrepresentation, we weighted each pair by the inverse of the distance between the roles (i.e. the distance as visualized in Figure 3). This means that pairs of microroles that are far apart are given greater weight in the language comparison, and pairs of microroles that are close together are given less weight. The resulting metric on the 25 languages is visualized in the form of a hierarchical clustering in Figure 6.12 A few clusters of languages are indicated in this figure, referring to Appendix 2, where the distributions of the coding devices of all theindividual languages are provided. We essentially find a major division between languages that do not show any coding differentiation between overt agent and patient marking (illustrated above by Balinese), and languages that do have some 12.  For this hierarchical clustering we used the function hclust from the base package of the statistical environment R. Specifically, we used the complete method for the clustering here.

Identifying semantic role clusters and alignment types via microrole coexpression tendencies

such division. Within this second group, a large set of languages has an almost identical agent/patient opposition, which primarily represents an accusative alignment pattern, with intransitive subjects classified together with transitive agents (illustrated above by Bora). However, the languages between these two groups are the most interesting. Four languages show a tripartite structure, separating different marking for intransitive subjects, transitive agents, and transitive patients (illustrated in Figure 6 by Jaminjung). The other languages show different variants between these clear groups, for example Hoocąk (as seen above) showing a split of the intransitive subjects between agent and patient (compare the division between the agent and patient groups between Bora and Hoocąk to see this difference). 8. Conclusion Traditionally, the comparison of languages is said to presuppose a thorough language-particular analysis (e.g. Newmeyer 1998: §6.5). But language-particular analysis involving abstract notions such as mesoroles and macroroles (or even cross-linguistic grammatical relations such as subject and object) is highly problematic, as these abstract analytic notions are not really comparable across languages. In this chapter, we have shown that it is possible to arrive at a role clustering and an alignment typology without presupposing any language-particular analysis beyond the level of the individual verb, and without any subjective judgment of semantic similarity. We only made use of information on argument coding (flagging and indexing) for each microrole of 87 verbs in 25 languages from around the world, allowing us to measure the similarities between the microroles and thus arrive at mesorole-like clusters (Figure 3), and in a next step to show the approximate distribution of coding sets over the microroles. This yields an objective, quantitative alignment typology that is not based on any deeper languageparticular analysis.

References Bickel, Balthasar. 2007. Typology in the 21st century: Major current developments. Linguistic Typology 11. 239–251.  doi: 10.1515/LINGTY.2007.018 Campbell, Eric. 2013. Zenzontepec Chatino valency patterns. In Iren Hartmann, Martin Haspelmath & Bradley Taylor (eds.), Valency Patterns Leipzig. Leipzig: Max Planck Institute for Evolutionary Anthropology. (http://valpal.info/languages/zenzontepec-chatino) Comrie, Bernard. 1981. Language universals and linguistic typology: Syntax and morphology. Oxford: Blackwell.

41

42

Iren Hartmann, Martin Haspelmath and Michael Cysouw Cysouw, Michael. 2010. Semantic maps as metrics on meaning. Linguistic Discovery 8(1). 70–95. Cysouw, Michael. 2014. Inducing semantic roles. In: Silvia Luraghi & Heiko Narrog (eds.), Perspectives on semantic roles, 23–68. Amsterdam: Benjamins. Dixon, R.M.W. 1979. Ergativity. Language 55. 59–138.  doi: 10.2307/412519 Donohue, Mark & Søren Wichmann (eds.). 2008. The typology of semantic alignment. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199238385.001.0001 François, Alexandre. 2008. Semantic maps and the typology of colexification: Intertwining polysemous networks across languages. In Martine Vanhove (ed.), From polysemy to semantic change, 163–216. (Studies in Language Companion Series 106). Amsterdam: Benjamins. Furrer, Reinhard, Douglas Nychka & Stephen Sain. 2012. Fields: Tools for spatial data. R package. Haiman, John. 1974. Concessives, conditionals, and verbs of volition. Foundations of Language 11(3). 341–359. Hartmann, Iren, Martin Haspelmath & Bradley Taylor (eds.). 2013. Valency Patterns Leipzig. Leipzig: Max Planck Institute for Evolutionary Anthropology. (http://www.valpal.info) Haspelmath, Martin. 2003. The geometry of grammatical meaning: Semantic maps and cross-linguistic comparison. In Michael Tomasello (ed.), The new psychology of language: Cognitive and functional approaches to language structure, vol. 2, 211–242. Mahwah, NJ: Erlbaum. Haspelmath, Martin. 2010. Comparative concepts and descriptive categories in cross-linguistic studies. Language 86(3). 663–687.  doi: 10.1353/lan.2010.0021 Haspelmath, Martin. 2011. On S, A, P, T, and R as comparative concepts for alignment typology. Lingustic Typology 15(3). 535–567. Haspelmath, Martin & Iren Hartmann. 2014. Comparing verbal valency across languages. In Bernard Comrie & Andrej L. Malchukov (eds.), Valency classes: A comparative handbook. Berlin: De Gruyter Mouton. Kibrik, Aleksandr E. 1997. Beyond subject and object: Toward a comprehensive relational typology. Linguistic Typology 1(3). 279–346.  doi: 10.1515/lity.1997.1.3.279 Lehmann, Christian. 2006. Participant roles, thematic roles and syntactic relations. In Tasaku Tsunoda & Taro Kageyama (eds.), Voice and grammatical relations: In honor of Masayoshi Shibatani, 153–174. Amsterdam: Benjamins.  doi: 10.1075/tsl.65.10leh Malchukov, Andrej L. 2013. Alignment preferences in basic and derived ditransitives. In Dik Bakker & Martin Haspelmath (eds.), Languages across boundaries: Studies in memory of Anna Siewierska, 263–289. Berlin: De Gruyter Mouton. Newmeyer, Frederick J. 2005. Possible and probable languages: A generative perspective on linguistic typology. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199274338.001.0001

Newmeyer, Frederick. 1998. Language form and language function. Cambridge, MA: MIT Press. Nichols, Johanna. 2008. Why are stative-active languages rare in Eurasia? A typological perspective on split-subject marking. In Mark Donohue & Søren Wichmann (eds.), The typology of semantic alignment, 121–140. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199238385.003.0005

R Development Core Team. 2012. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. http://www.R-project.org/. Van Valin, Robert D., Jr. 2005. Exploring the syntax-semantics interface. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9780511610578

Identifying semantic role clusters and alignment types via microrole coexpression tendencies Wälchli, Bernhard. 2009. Motion events in parallel texts: A study in primary-data typology. Bern: University of Bern Habilitationsschrift. Wälchli, Bernhard & Michael Cysouw. 2012. Lexical typology through similarity semantics: Toward a semantic map of motion verbs. Linguistics 50(3). 671–710.  doi: 10.1515/ling-2012-0021

Appendix 1.  The 87 verb meanings and microroles Meaning label

Role frame

Typical context

APPEAR

S appears

A deer appeared (in the road).

ASK FOR

A asks (X) for Y

The boy asked his parents for money.

BE A HUNTER S is a hunter

This man is a hunter.

BE DRY

S is dry

The ground is dry.

BE HUNGRY

E is hungry

The baby is hungry.

BE ILL

E is sick

The little boy is sick.

BE SAD

E is sad

The little girl was sad.

BEAT

A beats P (with I)

The boy beat the snake with a stick.

BLINK

S blinks

I blinked (my eyes).

BOIL

S boils.

The water is boiling.

BREAK

A breaks P (with I)

The boy broke the window with a stone.

BRING

A brings P T

The girls brought the old lady flowers.

BUILD

A builds P (out of X)

The men built a house out of wood.

BURN

S burns

The house is burning.

CARRY

A carries T (to X)

The men carried the boxes to the market.

CLIMB

A climbs (up L)

The men climbed (up) the tree.

COOK

A cooks P

The woman is cooking soup.

COUGH

S coughs

The old man coughed.

COVER

A covers P (with X)

The woman covered the boy with a blanket.

CRY

S cries

The baby is crying.

CUT

A cuts P (with I)

The woman cut the bread with a sharp knife.

DIE

S dies

The snake died.

DIG

A digs (for) P

The women are digging for potatoes.

DRESS

A dresses P

The mother dressed her daughter

EAT

A eats P

The boy ate the fruit.

FALL

S falls

The boy fell.

FEAR

E fears M

The man feared the bear.

FEEL COLD

S is cold

I’m cold.

43

44 Iren Hartmann, Martin Haspelmath and Michael Cysouw FEEL PAIN

E feels pain in M

My arm is hurting. = I’m feeling pain in my arm.

FILL

A fills P (with X)

The girl filled the glass with water.

FOLLOW

A follows X

The boys followed the girls.

FRIGHTEN

A frightens P

The bear frightened the man.

GIVE

A gives T to R

We gave the books to the children.

GO

S goes somewhere (L)

The woman went to the market.

GRIND

A grinds P (with I)

The boy is grinding corn with mortar and pestle.

HEAR

E hears M

The boy heard the bear.

HELP

A helps X

I helped the boys.

HIDE

A hides T (from X)

The boy hid the frog from his mother.

HIT

A hits P (with I)

The boy hit the snake with a stick.

HUG

A hugs P

The mother hugged her little boy.

JUMP

A jumps

The girl jumped.

KILL

A kills P (with I)

The man killed his enemy with a club.

KNOW

A knows P

The girl knew the boy.

LAUGH

S laughs

The little girl laughed.

LEAVE

A left L

The boy left the village.

LIKE

E likes M

The boy liked his new toy.

LIVE

S lives somewhere (L)

The old people live in town.

LOAD

A loads T (onto L)

The farmer loaded hay onto the truck. = The farmer loaded the truck with hay.

LOOK AT

A looks at P

The boy looked at the girl.

MAKE

A makes P (out of X)

The men made a house out of wood.

MEET

A meets X

The men met the boys.

NAME

A name X (a) Y

The parents called the baby Anna.

PEEL

A peels (X off) P

The boy peeled the bark off the stick.

PLAY

S plays

The child is playing.

POUR

A pours T somewhere (L)

The man poured water into the glass.

PUSH

A pushes P (somewhere)

The boy pushed the girl (into the water).

PUT

A puts T somewhere (L)

I put the cup onto the table.

RAIN

(it) rains

It rained yesterday.

RECEIVE=GET R gets T (from X)

The woman got a letter from an admirer.

ROLL

A rolls

The ball is rolling.

RUN

A runs

The horse is running.

SAY

A says “…”( to X)

They said “no” to me.

SCREAM

S screams

The man screamed.

Identifying semantic role clusters and alignment types via microrole coexpression tendencies SEARCH FOR

A searches for X

The men searched for the women.

SEE

E sees M

The man saw the bear.

SEND

A sends T (to X)

The girl sent flowers to her grandmother.

SHAVE

A shaves (his beard/hair)

The man shaved his beard/cut his hair

SHOUT AT

A shouts at X

The woman shouted at the children.

SHOW

A shows T (to R)

The girls showed pictures to the teacher.

SING

S sings

The boy sang (a song).

SINK

S sinks

The boat sank.

SIT

S sits somewhere (L)

The children sat on the floor.

SIT DOWN

S sits down (somewhere (L))

The children sat down on the bench.

SMELL

E smells M

The bear smelled the boy.

STEAL

A steals P (from X)

The boys stole apples from their neighbor.

TAKE

A takes P (from X)

The man took the money from his friend.

TALK

A talks (to X) (about Y)

The girl talked to the boy about her dog.

TEACH

A teaches R T

The teacher taught the children English.

TEAR

A tears P (from X)

The girl tore the page from the book.

TELL

A tells (X) Y

The girl told the boy a funny story.

THINK

A thinks about X

The girl thought about her grandmother yesterday.

THROW

A throws T somewhere (L) The boy threw the ball into the window.

TIE

A ties P (to L) (with I)

The man tied the horse with a rope to the tree.

TOUCH

A touches P (with I)

The boy touched the snake with a stick.

WANT

A wants X

The bear wanted this fish.

WASH

A washes P

The mother washed the baby.

WIPE

A wipes T (off L)

The girl wiped crumbs off the table.

45

46 Iren Hartmann, Martin Haspelmath and Michael Cysouw

Appendix 2.  Distribution of all language-specific coding devices, presented in 5 groupings (1) No Agent/Patient marking Jakarta.Indonesian Jakarta.Indonesian

Mandarin Mandarin

Balinese Balinese

zero zero

Mandinka Mandinka

Yoruba Yoruba

zero zero NP+la NP+la Loc Loc NP+ye NP+ye

Nluu Nluu

zero zero ng+la ng+la nla+NP nla+NP NP-dat NP-dat

zero zero aji+NP aji+NP ke+NP ke+NP uli+NP uli+NP

zero zero ba+NP ba+NP zai+NP+loc zai+NP+loc

zero zero ke+NP ke+NP

Xârâcúú Xârâcúú

zero zero fi+NP fi+NP ní+NP ní+NP LOC LOC sí+NP sí+NP

zero zero ngê+NP ngê+NP xù+NP xù+NP ti/tō+NP ti/tō+NP taa+NP taa+NP

Identifying semantic role clusters and alignment types via microrole coexpression tendencies (2) Agent/Patient marking Even

Mapudungun

NP-nom V.subj V.subj NP-nom NP-acc NP-acc NP-loc NP-loc NP-instr NP-instr NP-abl NP-abl NP-all NP-all NP-dat NP-dat

Bora

Italian

V.subj V.subj zero zero a+NP a+NP con+NP con+NP in/a/su+NP in/a/su+NP da+NP da+NP

V.subj V.subj V.obj V.obj NP+mew NP+mew zero zero

Arabic

V.subj V.subj NP-nom NP-nom NP-acc NP-acc bi+NP-gen bi+NP-gen min+NP-gen min+NP-gen ’ilâ+NP-gen ’ilâ+NP-gen ‘an+NP-gen ‘an+NP-gen

zero zero NP-ac NP-ac NP-adl NP-adl NP-abl NP-abl NP-instr NP-instr

Zenzontepec.Chatino

Yaqui

Ainu

zero zero NP-acc NP-acc NP-loc NP-loc NP-dir NP-dir NP-instr NP-instr

Sliammon

subj.V subj.V obj.V obj.V instr instr loc/all/abl loc/all/abl dat dat

zero zero jiʔį``+NP +NP jiʔį lóʔó+NP lóʔó+NP

(3) Agent/Patient with larger extend of Patient Bezhta

NP-abs & absAgr.V NP-erg NP-lat NP-gen1 NP-sup NP-instr NP-poss

V.sub V.sub V.obj V.obj obl-NP obl-NP

Hoocąk

act.V und.V -eeja/LOC unmarked NP

47

48 Iren Hartmann, Martin Haspelmath and Michael Cysouw (4) Tripartite

Alaskan.yupik Alaskan.yupik

Chintang Chintang

NP-rel V.subj NP-rel NP-absV.subj V.obj NP-abs V.subj V.obj NP-abs NP-abs V.subj NP-abm NP-abm NP-all NP-all NP-perl NP-perl

Jaminjung Jaminjung

NP-erg & V.subj [S] NP-erg V.subj [S] NP-abs & V.obj NP-abs V.subj V.obj [S] NP-abs NP-abs V.subj [S] NP-erg NP-erg NP-loc NP-loc NP-abs NP-abs

Nen Nen

NP-erg sub.V NP-erg obj.V sub.V NP-abs NP-abs subj.V obj.V NP-abs NP-abs subj.V NP-datV=pro.dat NP-datV=pro.dat NP-erg/instr NP-erg/instr NP-abs NP-abs NP-loc/all NP-loc/all NP-all NP-all

NP-abs & obj.V NP-abs & & V.subj obj.V [S] NP-erg NP-erg NP-abs & & V.subj V.subj [S] NP-abs & V.subj NP-aBS NP-aBS NP-dat NP-dat NP-dat & obj.V NP-dat & obj.V

Identifying semantic role clusters and alignment types via microrole coexpression tendencies 49 (5) Slightly tripartite

Icelandic

Ket

NP-nom & V.subj NP-acc NP-dat á+NP-acc meö+NP-dat af+NP-acc viö+NP-acc í+NP-dat um+NP-acc

subj.V obj.V NP-dat NP-abl NP-inst zero

Sri.Lanka.Malay

zero NP-acc NP-dat NP-abl

Semantic role clustering An empirical assessment of semantic role types in non-default case assignment Balthasar Bickel*, Taras Zakharko*, Lennart Bierkandt† and Alena Witzlack-Makarevich‡ *University of Zurich / †University of Jena / ‡University of Kiel

This chapter seeks to determine to what extent there is cross-linguistic evidence for postulating clusters of predicate-specific semantic roles such as experiencer, cognizer, possessor, etc. For this, we survey non-default case assignments in a sample of 141 languages and annotate the associated predicates for cross-linguistically recurrent semantic roles, such as ‘the one who feels cold’, ‘the one who eats sth.’, ‘the thing that is being eaten’. We then determine to what extent these roles are treated alike across languages, i.e. repeatedly grouped together under the same non-default case marker or under the same specific alternation with a non-default marker. Applying fuzzy cluster and NeighborNet algorithms to these data reveals cross-linguistic evidence for role clusters around experiencers, undergoers of body processes, and cognizers/perceivers in one- and two-place predicates; and around sources and transmitted speech in three-place predicates. No support emerges from non-default case assignment for any other role clusters that are traditionally assumed (e.g. for any distinctions among objects of twoargument predicates, or for distinctions between themes and instruments).

1. Introduction Apart from default or canonical case assignments, such as the assignment of accusative case to the most patient-like argument of transitives, many and perhaps most languages show alternatives in the form of non-default or non-canonical assignments for specific sets of predicates, e.g. the accusative on arguments of experience-denoting intransitives (as in German mich friert ‘I am cold’). It is often assumed that non-default marking of this kind does not occur at random. Indeed, several hypotheses and theories have been put forward that seek to predict the way in which semantic types of predicates associate with non-default case

doi 10.1075/bct.88.03bic 2016 © John Benjamins Publishing Company

52

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

assignments across languages (e.g. Tsunoda 1985, 2004, Onishi 2001, Haspelmath 2001, Malchukov 2005). However, all these hypotheses assume that predicates with non-default case assignments fall into natural semantic types such as experience, motion, uncontrolled event, etc. In other words, it is assumed that all lexical tokens of predicates, i.e. items like the German verb frieren or the English predicate expression be cold, can be successfully mapped into more general and more abstract classes like experience. A prominent correlate of this assumption is that predicate-specific argument roles cluster into more general and more abstract argument types — “role complexes”, as the editors of this volume call them — such as experiencer, theme, or instrument etc., and that these clusters are significantly similar across languages. This assumption is controversial. It is usually debated in terms of a choice between theoretical frameworks, e.g. by appeal to the efficiency and elegancy in describing general patterns of case assignment (e.g. capturing which intransitive verbs assign accusative rather than nominative case in German) or constructional constraints beyond case (e.g. in auxiliary choice or participle formation). Some theories assume role types (e.g. Lexical-Functional Grammar; Bresnan & Kanerva 1989, Butt 2008, Dalrymple & Nikolaeva 2011), others reject them (e.g. Role and Reference Grammar; Van Valin & Wilkins 1996, Van Valin 2005). In this chapter we want to turn the debate into an empirical one. Based on a typological database, we assess the empirical evidence for role clusters, asking to what extent non-default case assignment suggests natural and cross-linguistically relevant clusters: is there cross-linguistic evidence that non-default case assignment indeed systematically carves out, say, experiencers and themes as general types among the sole argument of one-place predicates? Is there evidence for carving out, say, perceiver and agent types among the most agent-like argument of two-place predicates (such as see vs. hit, etc.)? Is there any evidence for such type distinctions as recipients vs. spatial goals among the non-moving argument of three-place predicates (e.g. give vs. put)? We start by annotating case frames for predicate-specific roles (e.g. ‘the one who feels cold’, ‘the one who sees sth.’, ‘the one that gets hit by so.’, etc.) that recur across languages and that can be reasonably identified by translational approximation. We limit our attention to non-default case assignment (and non-default case alternations), assuming that defaults have no semantic specification of their own and cover everything that is not covered by non-default cases (or alternations). We then examine on a typological database to what extent predicate-specific semantic roles are grouped together by the same non-default cases (or alternations) in each language and derive from this a measure of dissimilarity of the roles across languages. The resulting dissimilarity matrix is then mined for statistical clusters, applying algorithms for fuzzy clustering (Kaufman & Rousseeuw 1990) and NeighborNet (Bryant & Moulton 2004, Huson & Bryant 2006) analysis. Any



Semantic role clustering

resulting cluster of predicate-specific roles is potentially indicative of cross-linguistically relevant role complexes. In the following, we first explain our notions of non-default case assignment and generalized argument classes (Section 2). Section 3 explains our database and the way we developed the cross-linguistic annotations of predicate-specific semantic roles. Section 4 describes the data-mining algorithms we used. Results of these are then presented in Section 5 and discussed in Section 6 in the light of expectations from the literature. The final section summarizes our findings. 2. Non-default case assignment and generalized argument classes Many languages exhibit diverse possibilities of case assignment.1 This can be illustrated with the following examples from Chechen (ISO639.3:che; NakhDaghestanian; Zarina Molochieva, p.c.). The clauses in (1) show that the sole argument of a one-place predicate can be in the absolutive, the dative, the ergative, and the allative case. (2) shows a selection of possibilities available for the arguments of two-argument predicates:2 (1) a. b. c. d.

so ohw-v-uuzhu-u. 1s.abs down-V-fall-prs ‘I fall down.’ suuna j-ouxa j-u. 1s.dat J-hot J-be.prs ‘I am hot.’ as jouxarsh tyyxi-ra. 1s.erg cough hit-witnessed.pst ‘I was coughing.’ soega nir qiett-a. 1s-all diarrhea strike-prf ‘I’ve got diarrhea.’

(2) a. as wazh b-u’-u. 1s.erg apple(B).abs B-eat-prs ‘I eat apples.’

1.  We use the term ‘case assignment’ in the broad sense of a paradigmatic contrast in the shape of noun phrases that distinguishes their roles as arguments of a predicate, including affixes, tone oppositions, adpositions, particles, morphological zeros in opposition to overt devices, etc. 2.  Glossing follows the Leipzig Glossing Rules; ‘V’, ‘J’, and ‘B’ denote genders.

53

54

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

b. c.

so hwo-x taxana qiet-a. 1s.abs 2s-lat today meet-prs ‘I meet you today.’ suuna Zaara j-iez-a. 1s.dat Zara(J).abs J-love-prs ‘I love Zara.’

Clearly, case assignment is sometimes not an isolated phenomenon but is part of a larger constructional choice. In (1d), for example, the allative is conditioned by the fact that the predicate is not expressed by a simple stem but instead by a complex lexicalized expression that involves the allative-assigning verb stem qiett- ‘strike’. Strictly speaking, then, the lexical entry ‘have diarrhea’ associates with the entire complex construction ‘allative+qiett-’ and not just with the allative declension form. In this chapter, we gloss over this complication for the following reason: Our interest lies in whether or not the roles that are licensed by various cross-linguistically identifiable lexical meanings (such as the role of the single argument of ‘have diarrhea’) are treated alike or not in a language, and whether there are systematic patterns behind this treatment across languages. For this question, it does not matter if a specific predicate meaning associates with a simple case choice or with a complex constructional choice of case and complex predicate structure at the same time. This difference is as irrelevant to our question as the difference between a case choice that affects only a simple suffix and one that involves some complex expression consisting of, say, a declension form and an adposition.3 As we will explain further below, we base our analysis of roles exclusively on the semantics of lexical entries (where Chechen nir qiett- licenses a single argument S just like English ‘have diarrhea’) and not on the formal shape of these entries (where one can argue about the transitivity of the expressions). It is commonly assumed that some types of case assignment represent the basic or canonical choice and others a non-basic, non-canonical choice. In Chechen, for example, one would consider the absolutive in (1a) and the ergative-absolutive frame in (2a) to be the basic choices. The range of individual predicates in basic case frames is typically open-ended, with no specified semantic limits. Open-ended classes of this kind are difficult to survey across languages because sufficiently rich dictionaries are scarce.

3.  In addition, we note that it can be very difficult to decide whether a specific case assignment is motivated by some sub-structure of the lexical predicate. The answer will often depend on the precise etymology of the expression and on the question to what extent speakers still have access to this sub-structure.



Semantic role clustering

One way out of this problem is to proceed with an a priori list of universal predicate meanings (like ‘eat’, ‘have diarrhea’, etc.) whose case assignments can then be catalogued for every language.4 Like all onomasiological (denotationbased, stimuli-based) approaches, this procedure allows easy comparison, but the pre-selection of predicate meanings brings with it the risk that the results are in part pre-determined. For example, it makes a difference for role clusters among intransitives how many different verbs of body functions (e.g. ‘belch’) there are, how many experience-related verbs (e.g. ‘be cold’) there are, etc.: if there are more experience-related meanings than body-function meanings in a list, evidence from case assignment patterns related to experiences weighs more than evidence related to body functions in cluster analyses, and this artificially favors the detection of experiencer clusters over role clusters related to body functions. While these problems can be kept somewhat at bay by enlarging lists of surveyed meanings and by trying to avoid Eurocentrism when compiling them, we explore an alternative approach here, concentrating exclusively on non-basic case assignment. The predicates associated with non-basic case assignment have the advantage that they are positively characterized by lists of verbs (which are also typically retrievable in grammars because one needs to say when the relevant nonbasic cases show up). Basic case assignment patterns, by contrast, can be expected to apply to open-ended lists of verbs, with an equally open diversity of meanings. Lists of verbs assigning non-basic cases can be readily surveyed and compared without any a priori assumptions about what to expect. But this approach is not without problems either. The most pressing one is how one can in fact distinguish basic from non-basic case frames. There are basically two classes of approaches to this, each of them replacing the intuitive notion of a basic choice in case assignment by more concrete concepts that can be better operationalized. In one approach, the notion of a basic choice is replaced by that of canonical arguments, or, more precisely, notions of canonically intransitive, canonically transitive, and canonically ditransitive argument frames. Canonicity is in turn established on the basis of a range of morphosyntactic, or semantically-grounded morphosyntactic, criteria so that, for instance, only the frames with accusatively-marked objects, or only with accusatively-marked and affected patient objects, or only with objects which can be promoted to subjects through passivization and which denote affected patients, are considered canonically transitive (cf. e.g. Onishi 2001, relying on Dixon 1994, but also more generally any research relying on notions like “quirky subjects”, “oblique objects”, etc.). 4.  This is the approach taken by the Leipzig Valency Class Project (Comrie & Malchukov In press, http: //www.eva.mpg.de/lingua/valency) and the Valency Project at the Russian Academy of Sciences, St. Petersburg (Say 2011).

55

56

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

In another approach, the notion of basic choice in case assignment patterns is grounded in prototypical predicate meanings. The choice is then based on pre-established notions of what would be prototypical representatives of one-, two-, and three-argument predicates, e.g. one would define predicates meaning something like ‘kill’ or ‘break’ as the prototypical representatives of two-place predicates (in the spirit of Comrie 1981) and take the case frame of these predicates to be basic. The first approach has been criticized for mixing semantic and syntactic criteria that are not strictly comparable across languages (Haspelmath 2011), and we do not adopt this approach here for this reason. The second approach does not suit the purpose of our investigation because it builds into a theoretical assumption what we want to explore empirically: the approach assumes a priori that across languages, predicates fall into at least two basic semantic types or clusters, a prototypical one including (in the case of two-argument predicates) meanings like ‘kill’ or ‘break’, and a non-prototypical one including meanings like ‘love’ or ‘see’. This may well be the case, but if so, we expect it to emerge empirically from a cluster analysis. Therefore, we need an alternative approach: we approach the notion of basic case frames with the idea of default case frames, i.e. case frames that are assigned when there is no other case specification in the lexical entry of a predicate. Default case frames in this sense are expected to be licensed by predicates that form an open class, and thereby by whatever class has the largest number of members in the lexicon and is most productive.5 For Chechen, this criterion establishes the predicate classes in (1a) and (2a) as the largest and most productive ones, and therefore their case frames, i.e. 〈abs〉 and 〈erg, abs〉, as the default. Any case assignment that deviates from these patterns is considered as the non-default frame of one-argument and two-argument predicates, respectively. For non-default frames, one needs to specify case assignments in the lexical entry of the predicate, e.g. one needs to stipulate that Chechen qiet- ‘meet’ assigns an 〈abs, lat〉 frame (as shown by (2b) above). This approach allows uncontroversial identification of default and non-default case frames among one-argument and two-argument predicates. The situation might at first sight seem less straightforward for split-S or “active” languages, which according to some analyses possess two equally substantial and equally 5.  Although in principle, class size and productivity might be in conflict, in practice they seem to coincide in determining one most common predicate class. For instance, Barðdal found a correlation between syntactic productivity and the size of predicate classes in the lexicon, as well as between the size of predicate classes and text frequency in Icelandic (Barðdal 2008: 60, 172f.). As the information on the number of predicates belonging to a particular lexical predicate class is relatively readily available for the languages surveyed, we chiefly rely on this criterion in our survey.



Semantic role clustering

salient sets of one-argument predicates (e.g. Comrie 2005). However, in each language of our sample, one of the one-argument classes clearly dominates the other in terms of the number of predicates.6 Problems arise, however, with three-argument predicates. As Malchukov et al. (2010) observe, three-argument predicates tend to be substantially less frequent in the lexicon than other predicates, and the small sets one finds are often very heterogeneous. In line with this observation, many languages in our sample have no clear default class of three-argument predicates. This is so, for example, in Tsamai (ISO639.3:tsb; Cushitic; Sava 2005). There are two case frames for three-argument predicates — 〈nom, acc, loc〉 and 〈nom, acc, abl〉 — and none of them outranks the other to such an extent that it could be taken to be the default choice. In such cases, we treated all classes as if they were non-defaults. Once non-default case assignment is identified in each language, the question arises of how one can compare the way in which non-default case assignment is associated with predicate-specific semantic roles. It would not be helpful to compare cases across just any kind of predicate-specific semantic roles because there is a fundamental difference between, say, the two roles licensed by two-place predicates and the three roles licensed by three-place predicates. For example, one would expect more predicate-specific semantic roles related to spatial transfer in three-place predicates than in two-place predicates, and this already imposes a limit on the kinds of clusters that can be found. Also, random comparisons of roles across predicates can create patterns that yield no real insight: if, say, the ‘what is eaten’ argument of ‘eat’ and the ‘one who meets’ argument of ‘meet’ happen to be assigned the same case in a language (e.g. absolutive, as in Chechen in (2)), one would not infer that these roles have something in common in their semantics. Therefore, comparisons of predicate-specific semantic roles are best based on a tertium comparationis that keeps general classes of arguments, such as the two arguments of ‘eat’, distinct. Traditionally, such classes are kept distinct on the basis of grammatical relations, and role clusters are then sought separately within ‘subjects’ and within ‘objects’. However, since grammatical relations incur well-known problems of cross-linguistic comparability due to their language-specific and constructionspecific nature (cf. Dryer 1997, Croft 2001, Bickel 2011, among many others), we proceed in a different way. We distinguish between argument classes first by numerical valence: the sole argument of one-argument predicates, the two arguments 6.  In a more extensive survey, Witzlack-Makarevich (2011) indeed finds no single language where case assignment would ever split one-argument predicates into classes that differ by a margin of less than 10% in their sizes. For case studies, see for example, Merlan (1985), Holisky (1987), or Pustet (2002). For a comparative lexicon investigation, see Nichols (2008).

57

58

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

of two-argument predicates, and the three arguments of three-argument predicates. The sole argument of one-argument predicates is symbolized as S. The arguments of two- and three-argument predicates are then distinguished on the basis of cross-linguistically viable lexical entailment properties (following a line of research going back to Dowty 1991):7 (3) Lexical entailments defining generalized argument classes a. A vs. P: A accumulates more lexical entailments than P on the following properties: – causing an event (e.g. A hits P, A kisses P, A goes to P, A meets P) – volitional (e.g. A hits P, A kisses P) – sentient (e.g. A sees P, A looks at P, A loves P, P pleases A) – independently existing (e.g. A bakes P, A makes P) – having possession over another participant (e.g. A has P, P belongs to A) b. G vs. T: G accumulates more lexical entailments than T on the following properties: – stationary relative to movement of another participant (e.g. A gives T to G, A loads T onto G, A covers G with T, A cuts G with T) – receiving or being exposed to an experience (e.g. A shows T to G, A tells T to G)

For instance, kiss — as in Lisa kissed Tom — entails that Lisa is causing the event of kissing, behaves volitionally, is sentient of this event, and exists independently of kissing. Tom exists independently of kissing, but he is neither causing this event, nor is he behaving volitionally, nor is he necessarily sentient of it (e.g. if he is in coma or asleep). Thus, the kisser accumulates more of the lexical entailments defining the A argument than the kissee and qualifies as the A argument, while the kissee is then necessarily the P argument. Note that ‘A’ in (3) stands for the A argument class of two-argument predicates only. Three-argument predicates have an ‘A2’ argument (Bickel & Nichols 2009, Bickel 2011), and this is distinguished from T and G in the same way as A is distinguished from P arguments. The main reason for distinguishing between A and A2 is that the kinds of meanings covered by the A2 role of three-argument predicates are different from those of two-argument predicates (compare e.g. ‘the one who 7.  The definition of the G vs. T contrast deviates from the one in Bickel et al. (2010) and Witzlack-Makarevich (2011), partly in response to an insightful critique in Schikowski (2013). For a recent comparison of various approaches to thematic roles including some critique of Dowty’s approach, see Croft (2012). Note that we differ from Dowty (1991) in that we set up A and P not as independent but as relative prototypes, following Primus (1999, 2006).



Semantic role clustering

sends’, ‘the one who gives’ among three-argument predicates with ‘the one who perceives’, ‘the one who hits’ etc. among two-argument predicates). This difference often leaves no reflex in case marking and most languages treat A2 in exactly the same way as the A argument of two-argument predicates (but see Bickel & Nichols 2009). However, with non-default cases, the difference between A and A2 matters more commonly. For example, emotion-related dative experiencers (‘be afraid of something or someone’) are typically not found among three-argument predicates but only among two-argument predicates. Note that the generalized argument classes, as we define them here, are strictly independent of their morphosyntactic realization and thereby of any association with a specific case frame (or indeed, of any association with a case frame plus some formal sub-structure in the predicate, as discussed in Section 2). This makes it possible for these classes to serve as a tertium comparationis when comparing the cases that specific predicates associate with. Applying our definitions of generalized classes to the Chechen examples above, we obtain three non-default case frames for the data in (1), viz. 〈S-dat〉, 〈S-erg〉, and 〈S-all〉, and two non-default case frames for the data in (2), viz. 〈A-abs, P-lat〉, and 〈A-dat, P-abs〉. While they are fully independent of morphosyntax, the argument classes in (3) are not fully independent of semantic roles and role clusters: they presuppose that agents and patients, or goals and themes are different from each other, respectively, and could never end up in the same cluster. This is a limitation to our approach. However, the damage is minimal because, as noted above, clusters combining agents and patients, or goals and themes, would be difficult, perhaps even impossible, to interpret semantically, i.e. they would not yield a consistent role type. Also, the way we set up the classes still leaves much room for role clusters within each class, all of them with a realistic chance for the kind of semantic homogeneity that theories of semantic roles predict, e.g. an experiencer vs. a theme role among S and A arguments, or a recipient vs. spatial goal role among G arguments. Equipped with argument classes and a definition of non-default case assignment, we can now compare case assignment patterns for each generalized class: are there any cross-linguistic trends in how non-default case assignments cluster predicate-specific roles in the class of S arguments, in the class of A arguments, T arguments, etc.?

59

60 Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

3. Data We surveyed a sample of 141 languages and coded their case systems. The sample is geographically widespread, covering the whole world (see the map in Figure 1).8 The sample is also genealogically balanced along the lines proposed by Dryer (1989): we excluded non-default case assignment rules that exactly replicate rules in other languages within the same major branch of a family (in the sense of Nichols et al. 2013). This rules out signals being occasionally inflated by larger families, but it also means that our study is only exploratory at this point and cannot assess the diachronic dynamics that lead to the patterns we find.9 In each language of the sample, we identified each set of predicates that is associated with a specific non-default case frame, or a specific case frame alternation involving a non-default marker (such as alternations conditioned by tense or aspect or by referential properties in the form of differential object marking, or by perspectivization choices as in ‘spray/load’-type alternations etc.). We annotated each resulting set for the range of meanings in its members, approximated by English translation equivalents as found in or inferred from available lexical resources in dictionaries and grammars. We only included meanings that are specific to a predicate: for example, when the description of a non-default case frame only vaguely refers to “verbs of transfer”, without further specifications (e.g. whether this includes verbs like ‘send’ or ‘tell’), we did not include this as the meaning ‘transfer’. There is no doubt that our data collection and annotation misses relevant predicates in some languages, e.g. because the predicates were simply forgotten when grammar authors set up lists. Translations can include errors as well. Specifically, we, or our sources, may have occasionally missed semantic compositionality and inadvertently analyzed as a single complex predicate (e.g. ‘cough’) what really should be analyzed as a two-place predicate (‘throw out particles’, with particles as a referring expression).10 As a result, it is clear that our data is noisy to some extent. However, if semantic role clusters are cross-linguistically real and relevant, we expect them to leave statistical signals in our data despite all this noise — perhaps not very strong and clear, but at least detectable signals. At any rate, the raw data that we obtained in this way consists of tables as in Table 1 for Chechen.

8.  The database and the statistical analyses reported below are available as an online appendix at DOI 10.1075/sl.38.3.03bic.additional. 9.  For a diachronic study, one would have to have much denser sampling in large families. 10.  Our analyses can be inspected and checked in the online appendix.



Semantic role clustering

Figure 1.  Languages in the sample (N = 141)

Based on tables like these, we extracted the meanings associated with each class-case combination, i.e. with each combination of an argument class (e.g. S) with a specific case or case alternation (e.g. dat or erg/abs). For example, for the class of S arguments we would collect all predicate-specific roles covered by the same case or case alternation. Applied to Table 1, this results in one set including the roles of ‘the one who has a fever’, ‘the one who feels hot’, and ‘the one who feels good’ (covered by the class-case combination S-dat), another set containing the roles of ‘the one who blinks’, ‘the one who loses hope’, ‘the one who feels insulted’, ‘the one who coughs’, ‘the one who sins’, and ‘the one who grows’ (all assigned ergative case in the continuous and absolutive in the non-continuous aspect, i.e. the class-case combination S-erg/abs), and finally, a set consisting of ‘the one who starts coughing’ and ‘the one who has diarrhea’ (the class-case combination Table 1.  Chechen non-default case frames or case frame alternations and the meanings of the associated predicates. (The erg/abs-alternation is conditioned by aspect; see Molochieva 2010). Case frame/alternation

Meanings

〈S-dat〉

have fever, feel hot, feel good

〈S-erg/abs〉

blink, lose hope, be insulted, cough, sin, grow

〈S-all〉

start coughing, have diarrhea

〈A-dat, P-abs〉

see, hear, like, love, remember, lack

〈A-abs, P-all〉

look at, wait for, scold

〈A-abs, P-lat〉

depend on be proud of, fear

〈A-erg/abs, P-dat〉

look after, surround, lock, help

〈A-erg/abs, P-lat〉

abuse, forgive

〈A2-erg/abs, T-abs, G-all〉

say, send, loan, ask, remind

〈A2-erg/abs, T-lat, G-abs〉

load, support

〈A2-erg/abs, T-ins, G-abs〉

fill, smear, cover, congratulate

61

62

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

S-all). To keep the description manageable we refer to the predicate-specific roles in the following simply by the argument class and the predicate meanings, e.g. ‘S-have_fever’ stands for an S-class semantic role ‘the one who has a fever’, etc. The extraction of predicate-specific semantic roles with non-default case assignments is trivial for S-case combinations because each such combination is associated with a unique set of roles. For other arguments, the semantic roles need to be collected across case frames. For example, in Table 1, the absolutive-marked A occurs in two different case frames, viz. 〈A-abs, P-all〉 and 〈A-abs, P-lat〉. Therefore the class-case A-abs is associated with the set union of the meanings of these two frames: {A-look_at, A-wait_for, A-scold, A-depend_on, A-be_proud_of, A-fear}. When a class-case combination reflects the default, we do not analyze it further. For example, A-erg/abs in Chechen is the case alternation pattern that is associated with the default frame (cf. the data in (2a) above), and so we do not analyze the meanings covered by the class-case combination A-erg/abs. As a result of this, the meanings associated with the frames 〈A-erg/abs P-dat〉 and 〈A-erg/ abs P-lat〉 in Table 1 enter the analysis only with regard to the P-dat and P-lat class-case combinations, where they do not correspond to a default. This results in the role sets {P-look_after, P-surround, P-lock, P-help} as covered by the dative case and {P-depend_on, P-be_proud_of, P-fear, P-abuse, P-forgive} as covered by the lative case. A special situation arises when a language lacks a default case frame, which is (as noted earlier) fairly common among three-argument predicates. In these cases, we analyzed a class-case combination as a relevant (i.e. non-default) instance if (and only if) it does not extend across all case frames in three-argument predicates. We noted above (Section 2) the situation in Tsamai. In the two three-argument frames — 〈A2-nom, T-acc, G-loc〉 and 〈A2-nom, T-acc, G-abl〉 — the class-case combinations A2-nom and T-acc occur in both case frames, and so we exclude these two, assuming that they cannot give evidence of any semantic specification. Thus, Tsamai three-argument predicates enter the analysis only in terms of the G class, with specifications for the class-case combinations G-loc and G-abl. Once these analyses were performed, we searched through all sets of predicatespecific semantic roles that are defined by non-default case assignment and looked for recurrent items. We interpreted these items as proxies of cross-linguistically relevant roles (such as S-have_fever, A-love, P-fear, etc.). The question arises of what one would take as the minimum amount of recurrence for exploring crosslinguistic patterns in roles. Table 2 surveys how often roles recur across languages, i.e. how often they are mentioned in our database as being assigned a non-default case or undergo non-default case alternations. Given the skewing towards low N in Table 2, any cut-off point will have to be relatively low unless one wants to throw out most of the data. Since any choice here



Semantic role clustering

Table 2.  Number of predicate-specific roles that are assigned a non-default case or a nondefault case alternation N times N = 1

N = 2

N = [3,4]

N ≥ 5

Total

S

  61

16

16

 9

102

A

  88

33

18

24

163

P

116

57

46

70

289

A2

   1

 0

 0

 0

   1

G

  34

19

16

18

  87

T

  20

13

 9

 9

  51

would be arbitrary (say, requiring a role to be mentioned in at least 5 languages), we decided to take into consideration all roles that recur at least once (i.e. that are mentioned in at least two rules of non-default case assignment or non-default case alternation, N ≥ 2). This means that even a single recurrence of a pattern (e.g. that two languages assign S-have_fever and S-be_cold the same non-default case) is taken as a relevant signal for role clustering. However, we will take the amount of recurrence into account when interpreting results. Note that for the A2 class of arguments (in three-place predicates), only one language in our dataset shows a non-default class-case combination. This is Marathi (ISO639.3:mar; Indo-Aryan), where verbs of telling assign nominative instead of ergative in all tenses, whereas the default rule is to assign ergative in the past tense (Pandharipande 1997: 132). All other languages in our database assign A2 the same case throughout, across all three-place predicates. We therefore exclude the class of A2 arguments from further analyses. Requiring roles to recur at least once (i.e. N ≥ 2 in Table 2) leaves us with 41 predicate-specific roles in the class of S arguments, 75 in the class of A arguments (in two-place predicates), 173 in the class of P, 53 in the class of G, and 31 in the class of T arguments. 4. Methods Our goal is to identify the extent to which non-default case assignment targets similar or dissimilar predicate-specific semantic roles. For example, if non-default case assignment to S recurrently treated roles related to experiences (S-feel_cold, S-feel_sick, etc.) alike, we would interpret this as evidence for a cluster of experiencer roles in the class of S arguments. In order to derive a suitable measure for assessing whether semantic roles are treated similarly vs. differently, we first tabulate all roles and specify whether or

63

64 Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

not a given role is listed for each class-case combination in the database. Table 3 illustrates this for a selection of class-case combinations. For example, we observed above that the Chechen A-abs combination associates with the set of predicatespecific semantic roles {A-look_at, A-wait_for, A-scold, A-depend_on, A-be_ proud_of, A-fear}. In Table 3, this is represented as follows. Chechen A-abs (in the first column) gets 0 (absence) for A-love and A-hate since these semantic roles are not in the set, but 1 (presence) for A-fear and A-wait_for because these components are in the set. Other class-case combinations in Chechen and other Table 3.  Matrix of semantic roles Chechen

Djambarrpuyqu

Tsez

Nias

…

A-abs

A-nom

A-dat

A-abs

…

A-love

0

1

1

1

…

A-hate

0

0

1

1

…

A-fear

1

1

0

1

…

A-wait_for

1

0

0

0

…

…

…

…

…

…

…

languages show a different pattern. We then compute the pair-wise mean dissimilarity across rows in the matrix, using the Jaccard distance. The Jaccard distance is defined as the proportion of differences (i.e. with ‘0’ in one row and ‘1’ in the other) among all columns that do not contain ‘0’ in both rows. Discarding pairs with both ‘0’ (or blanks) is important because the absence of some non-default case assignment only means that the roles are ignored by the relevant assignment rule, not that they are treated alike. Table 4 shows the dissimilarity matrix that is derived from Table 3 by computing the Jaccard distance. Table 4.  Dissimilarity matrix of semantic roles A-love

A-hate

A-hate

1/3 = .33

A-fear

2/4 = .5

3/4 = .75

A-wait_for

4/4 = 1

3/3 = 1

A-fear

2/3 = .66

For example, comparing the rows with A-love and those with A-hate in Table 3 shows three columns with pairs that are not both zeros, i.e. the Djambarrpuyqu, Tsez, and the Nias column. One of these shows presence (1) for A-love and absence



Semantic role clustering

(0) for A-hate; two show presence in both. This results in a Jaccard distance of 1/3. In the comparison of the rows with A-love and A-fear, four (i.e. here, all) columns have at least one ‘1’ in both rows. In two columns (Nias and Chechen) values disagree and so the distance is 2/4. We compute such dissimilarity matrices separately for each of the five argument classes retained for analysis: S, A, P, G, and T. The resulting matrices are then mined for clusters. There are many techniques available for this, each with its advantages and disadvantages (cf. Kaufman & Rousseeuw 1990, Rokach 2010). In order to increase the chances of detecting patterns in the data we combined two methods: fuzzy cluster analysis and NeighborNets. The basic idea of fuzzy cluster analysis is not to partition the data, here predicate-specific semantic roles, into a categorical set of clusters, but to partition the data into clusters with fuzzy boundaries and centered around a prototype. This fits with common assumptions that, like other semantic categories, semantic roles are structured around prototypes. Fuzzy cluster analysis assigns each member of a cluster a membership coefficient between 0 and 1, indicating the probability of being a member in this cluster (so that k represents equiprobable membership in k clusters). The challenge is to optimize these membership coefficients in such a way that the coefficients minimize within-cluster dissimilarity and maximize betweencluster dissimilarity, given a number of clusters. Kaufman & Rousseeuw (1990) introduced an iterative algorithm to solve this optimization problem (implemented in R (R Development Core Team 2013) and made available by Maechler et al. 2005). As individual datasets allow for many ways of fuzzy clustering, one needs to determine the most adequate number of clusters (k). This is done by considering a reasonable range of clusters and selecting the one for further analysis that shows the clearest, “crispest” partition of items (here, semantic roles). The standard goodness-of-fit statistic for crispness of clusters is known as the Dunn statistic (or Dunn coefficient) which in its normalized form ranges from 1 (crisp, wellmotivated clustering) to 0 (no evidence for clustering, i.e. membership is always equiprobable across all clusters). The best solution is then defined as the one that combines the lowest number of clusters (k) with the highest Dunn statistic. We search for this optimum ‘by hand’, i.e. by inspecting plots of Dunn statistics against numbers of clusters (printed below, in the Results section) and choosing the optimum as the point where the Dunn statistic reaches its first peak or where it begins to flatten out (i.e. where higher numbers of clusters no longer improve the statistic). Crisper clustering can generally be obtained by lowering what is called the membership exponent (ME), which serves to weigh the membership coefficients during the algorithm. While yielding crisper signals, low exponents can impede the convergence of the algorithm (Kaufman & Rousseeuw 1990). We therefore chose the minimal exponent that allowed convergence across datasets.

65

66 Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

Fuzzy cluster analysis results in lists of sets and estimates of the degree to which each semantic role belongs to any one of these sets, with the prototype members having membership coefficients close to 1. What is missing from this method is a direct and easy-to-visualize estimate of the similarities and dissimilarities between clusters. These (dis‑)similarities are of interest for our study because they could reveal higherorder relations between role clusters. In order to derive estimates of cross-cluster relations, we also performed splits-graph analyses of the data, using the NeighborNet method as implemented in the SplitsTree package (Bryant & Moulton 2004, Huson & Bryant 2006). NeighborNet is an algorithm that solves a fundamental problem of visualizing dissimilarity (or distance) matrices like the one in Table 4: if we know the distances between 3 elements, we can draw the 3 locations on a plane without distorting the distances. If there are more elements and the distances vary freely, this requires more dimensions, defeating visualization on a plane. The NeighborNet algorithm basically solves the problem by splitting the paths between elements. We illustrate the application of this method on a small set of data in Table 4. The distance matrix in Table 4 cannot be plotted on a plane without distorting distances: it is geometrically impossible to place A-wait_for in such a way that it has a distance of 1 to A-love and A-hate, and that at the same time the fourth item A-fear has a distance of one half to A-love and three quarters to A-hate. The NeighborNet in Figure 2 solves the problem by splitting up paths in such a way that they represent the distances faithfully. This can be verified by adding path lengths and comparing them to the original matrix, allowing for rounding effects. For instance, to find the distance between A-wait_for and A-fear, one calculates the sum of the lengths of the edges in the path between these two elements: .457 + .125 + .084 = .666. This sum corresponds to the Jaccard distance in Table 4. A-hate .167

.042 A-love

A-wait_for

.457

.375 .125 .084 A-fear

Figure 2.  A NeighborNet visualization of the dissimilarity matrix in Table 4



Semantic role clustering

We produced NeighborNet representations of the dissimilarity matrices of each argument class and then superimposed the results of the fuzzy cluster analysis to display where the prototypes are of each cluster and how far away individual components are from the prototype. For this we use different colors for each cluster and let the saturation of the color show the membership coefficient, starting with full saturation for a membership coefficient of 1 and then shading into less saturated colors and reaching white where membership coefficients equal equiprobable membership (1/k). Not all cluster signals have the same weight in interpretation: if two roles are frequently treated alike by non-default case assignment, this is stronger evidence for the cluster than if the (dis)similarity assessment is based on fewer non-default case assignment rules because the roles are marked by default cases in most languages. In order to keep this information visible when interpreting the analyses, we scaled the font sizes according to the frequency range of roles displayed above in Table 2: if roles are assigned non-default cases in only two languages, we choose the smallest font, if they are assigned non-default cases in between three and five languages, the role label is printed in the second-smallest size, and if they occur in between six and ten languages, the second-largest font isued. The largest font size is reserved for roles that are assigned non-default cases in at least 11 languages.11 5. Results With a membership exponent of ME = 1.1, it was possible to obtain fuzzy cluster solutions for all dissimilarity matrices, except for the matrix attached to the P class of arguments. For the P class, the fuzzy cluster algorithm was unable to converge with ME ≤ 1.2, and with higher exponents, membership coefficients were all close to equiprobable membership (i.e. close to 1/k). We conclude that there is no appreciable clustering of semantic roles in the P class and exclude it from further analysis. This leaves us with the four classes S, A, G, and T. For these, Figure 3 plots the Dunn statistic as measures of crispness of the fuzzy clustering. Based on the plots in Figure 3, we performed fuzzy cluster analyses with 3 clusters for S, 6 clusters for A and G each,12 and 4 clusters for T because these represent the first local maxima or the point where the curve begins to flatten out. The results are shown in Figures 4 and 5. We first discuss the results for the ‘subject’ 11.  Detailed inspection of frequencies is possible in the online appendix. 12.  For A and G we also considered analyses with more clusters, but the NeighborNet analyses printed below show that any further clusters are so close to each other that the partition would seem an artifact of the method, with limited empirical interest.

67

68 Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich 2 3 4 5 6 7 8 S

A

2 3 4 5 6 7 8 G

T

D

0.8 0.6 0.4 0.2 2 3 4 5 6 7 8

2 3 4 5 6 7 8 Number of clusters

Figure 3.  Dunn statistic (D) per number of clusters for each argument class

classes (S and A) and then focus on the results for the ‘object’ roles (G and T) of three-argument predicates. 5.1 S and A classes of semantic roles The S class (top graph in Figure 4) shows a relatively clear separation between experiencer roles on the right-hand side of the graph (Cluster 1, colored red) and undergoers of body processes on the left-hand side (Clusters 2 and 3). The best exemplars of experiencer roles are located at the bottom of the graph (e.g. S-feel_like_laughing, S-feel_ashamed, S-feel_fear, with membership coefficients higher than .99), but they are closely followed by other roles that are distributed all over (e.g. S-feel_thirsty, S-be_surprised, S-feel_cold, etc., with membership coefficients higher than .98). From a conceptual point of view, no specific structures emerge within the experiencer cluster. For example, no division emerges between emotions and sensations, or between experiences that affect the whole body and experiences that are more specialized. The NeighborNet suggests a continuum within the cluster (so that path lengths between adjacent roles are shorter than between more distant roles), but as far as we can tell, there is no conceptual rationale motivating this. For undergoers of body processes, by contrast, the fuzzy cluster analysis shows a relatively clear subdivision: Cluster 2 (green) appears to combine processes that are more open to voluntary intervention and control than the processes in Cluster 3 (dark blue). The prototype roles (with membership coefficients higher than .99) for the better controllable processes are S-urinate, S-defecate, S-laugh, S-shed_ tears, and S-sleep. The cluster of less well controllable processes is centered around the roles S-shiver and S-fall. The roles S-belch, S-be_crazy, S-wake_ up, and S-grow also belong to the prototype range of less well controllable processes (membership coefficients above .99), but with data support from only two languages each (as shown by the small font size). While the subdivision between



Semantic role clustering 69

Figure 4.  NeighborNet and fuzzy clustering of predicate-specific roles in non-default case assignment to S (top) and A (bottom) arguments. (Roles are represented by predicates, but these are meant to refer to the respective arguments, e.g. break in the S class refers to ‘that which breaks’.)

70 Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

Figure 5.  NeighborNet and fuzzy clustering of predicate-specific roles in non-default case assignment to G (top) and T (bottom) arguments. (Labeling conventions as in Figure 4)



Semantic role clustering

less vs. better controllable body processes is well supported by the fuzzy cluster analysis (cf. the high Dunn statistic for a three-cluster analysis of the S class in Figure 3), the NeighborNet also makes clear that there is a continuum between the two, making up an overall cluster of undergoers of body processes. Note that this cluster also contains a number of predicates which one would traditionally analyze as involving themes (swim, grow, appear, fall, etc.), but there is no evidence that these constitute a special subcluster. The experiencer role in the S class is partially replicated in the A class, but the support for this is weaker (in line with generally lower crispness of the clustering, as shown in Figure 3): experiencers tend to be grouped as Cluster 1 (red, on the right-hand side of the bottom graph in Figure 4), centered around A-feel_ irritated_w, A-enjoy, and A-annoy (with membership coefficients above .99). However, other roles related to emotions and sensations are scattered across other clusters (e.g. A-be_tasty_to, A-fear, A-envy, etc.). Given the low numbers of mentionings (cf. the small font size), however, neither the experiencer cluster nor the exceptions to it should be accorded too much weight in the overall interpretation of the results. The evidence here is clearly weak. The A class shows a few other tentative clusters that are supported both by the NeighborNet and the fuzzy cluster analysis: one of these is Cluster 4 (purple, on the right-hand side of the graph), which combines A roles in cognition and perception, centered around the prototype A-see (with a membership coefficient of .98) and immediately followed by A-find, A-know, and A-hate (membership coefficient of .97). The cluster does not extend to what appears to be related roles such as A-think_about or A-believe, which more often group with other roles in our data. Another cluster with relatively good support is Cluster 6 (light blue, on the left-hand side of the graph), which mostly combines roles involved in transformations (A-transform, A-finish, A-make, A-plan, A-become and — not really fitting here, A-resemble, all with a membership coefficient higher than .99). Finally, Cluster 3 (dark blue, upper left) tentatively suggests a cluster of roles involved in communication, with the best representatives being A-listen_to (membership coefficient .97) and A-call_for (.96); A-wait_for shows a similar membership coefficient (.97), but the NeighborNet suggests that the role is also very similar to other roles outside Cluster 3. Also note that the ‘transformer’ and ‘communicator’ roles are supported only by relatively few languages and can only be taken as highly tentative. 5.2 G and T classes of arguments Like in the A class, the G class of arguments does not show very crisp clustering (cf. the low Dunn statistic in Figure 3), and this is confirmed by the NeighborNet which suggests similarly long distances between most roles (cf. the top graph in Figure 5).

71

72

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

But some trends in the analysis are detectable nevertheless. Cluster 1 (red, top right) appears to be clustered around source locations, specifically around G-get_from and G-buy_from (with membership coefficients of .99 and .98, respectively). Another potential trend is Cluster 2 (green, lower right), with goals of spatial transfer (G-give and G-send_to) at the center. However, the evidence for such a cluster is weak: even the central roles reach membership coefficients of only .90 and .89 and the NeighborNet suggests continuity with G class arguments of communication (say, ask, teach, explain to) and transaction verbs (sell, borrow from, lend to) in what is placed in Cluster 5 (yellow, just above Cluster 1) by the fuzzy cluster analysis. The only other cluster that is supported by closeness in the NeighborNet analysis is Cluster 4 (purple, top left in the graph). This cluster is centered around G-quarrel_with and G-pay_with_for (each with .99 membership coefficients), but it is unclear whether there are any systematic conceptual grounds that would motivate this cluster as distinct from the roles in other parts of the graph. The trend towards a transfer-based cluster around G-give and G-send_to is not paralleled by the data from T class arguments (cf. the bottom graph in Figure 5). T-give and T-send_to appear in Cluster 2 (green) but not as prototypes, and the NeighborNet analysis shows that they are at almost maximum distance from each other (T-give appears in the upper left area of the graph while T-send_to is placed at the bottom). What seems more relevant for the overall structure in the T class is a continuum between theme-of-motion meanings starting in the middle of the graph (Cluster 2, green, both upper and lower side) and ending on the right-hand side with roles covering material and things that become attached to something (Cluster 1 and 3, red and blue). The only apparent exception is T-hit (top side), but the conceptual distance is not too far in fact: an instrument that one uses for hitting becomes attached to an object if only perhaps briefly while, say, a cover or load tends to stay in place once it is applied. However, we note that the area of attached items does not cluster too well, spread as it is between two of the clusters suggested by the fuzzy cluster analysis. Also, themes of transfer motion appear on both the upper and lower side in the middle of the NeighborNet, showing more diversity than the fuzzy cluster analysis suggests by grouping roles in Cluster 2 (green). The only cluster that is consistently separate across both the Fuzzy Cluster and the NeighborNet analysis in the T class is Cluster 4 (purple, lower left side). This cluster is mostly about speech arguments (‘what is communicated’) and is centered around T-allow, T-teach, T-demand_of (all with membership coefficients above .99). Some roles do not fit, however, such as T-sell (which has a membership coefficient of .99 as well) and some verbs of saying are placed elsewhere (e.g. T-accuse_of, T-name).



Semantic role clustering

6. Discussion This study is based on surveying lists of predicates associated with non-default case assignment, as found in descriptive grammars and dictionaries. As noted in Section 3, this method of data collection necessarily brings with it a certain amount of uncertainty. However, there is evidence that, despite this, the data are sufficiently clear: we do find some clear clustering effects in some areas but not in others. If the data were too noisy or too skewed (e.g. because some predicate meanings were systematically underreported in grammars), one would have expected much fewer or much more (and crisper) clustering than we actually found. This suggests that the results allow further interpretation and comparison with traditional expectations about semantic role clusters. There are a few areas where our analysis confirms traditional expectations. The best fit with traditional role notions is found in the class of S arguments. Here, a broad distinction emerges between experiencers as opposed to undergoers of body processes, the latter subdivided into more vs. less controllable processes (e.g. urinate vs. shiver). The experiencer cluster is weakly supported by the A class of arguments as well. While the S class does not suggest any further subdivisions of experiencers into, say, emotions vs. sensations, the A class shows a separation of experiencers vs. cognizers/perceivers, although the signal here is not very strong. Another traditional role that can be detected is the spatial source role complex among G arguments. The other argument classes do not show clusterings reminiscent of traditional semantic role types. Most importantly, the analysis of the T class of arguments fails to support a clear distinction between instruments and themes. What emerges instead is a loosely structured continuum ranging from themes of transfer to attached items. While unexpected from the point of view of a priori theorizing, this result is in line with the observation that it is often difficult to tell apart instruments and themes. For example, the T argument of verbs like ‘to fill’ can be conceived both as a theme that is moved into a container or as an instrument with which one changes the state of the container. What seems more relevant is a notion of attachment, whereby items end up in an attached or otherwise fixed state. More generally, it seems that spatial transfer is less relevant for three-argument predicates than traditionally assumed. There is a cluster of goals of spatial transfer among G arguments (with the goals of give and send to at the center), but as noted above, the cluster is not well separated from verbs of communication (say) and transaction (sell). This can mean that, despite widespread assumptions, spatial transfer is not a prototypical concept for three-argument predicates. However, our finding can also mean that spatial transfer is only relevant as a prototype outside non-default case assignment. This second possibility would require

73

74

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich

that the opposite of non-default case assignment, i.e. ‘basic’ case frames, is not conceived of as a default in our sense (cf. Section 2), but that ‘basic’ case frames are conceived of as positively defined semantic categories, organized around prototypes. Evidence for such an interpretation cannot come from our study (since it is limited to non-basic case frames), but requires in-depth work on the semantic fine-structure of basic case frames in three-place predicates. Our study also suggests a number of role clusters that are not canonically assumed in theories of semantic roles. The most important concept here is that of speech/concept. A role cluster of communication items (e.g. T-allow or T-demand_of) emerges as a relatively well supported cluster in the T-class, and there is tentative and weak evidence for a cluster of ‘communicator’ roles in the A class (e.g. A-listen_to or call_for). In general, clustering is weaker among the A and G classes of roles than among the S and T classes. It seems that when languages assign non-default cases to A and G arguments, they do so in ways that cannot be easily predicted by predicate semantics. For the A class, this may perhaps be explained by assuming that case markers on A arguments mostly serve the discriminatory rather than indexing function of case (Comrie 1978, 1981, Song 2001), i.e. these case markers simply keep A noun phrases distinct from P noun phrases and do not index any semantic subgroups of roles. This explanation fits with the observation that the P argument class does not show any role clustering at all. However, the absence of statistical signals can also be caused by the limited size of our dataset and we do not want to push any explanation here too far. Also, we note that the distinction between discriminatory and indexing functions is not as straightforward as traditionally assumed (Arkadiev 2008). For the G class, one possible explanation comes from the fact that these roles are commonly associated with various locative cases and that locative cases (e.g. various subtypes of allatives) tend to easily extend and reduce their semantic ranges over time for reasons that have little to do with predicate semantics and more with the conceptualization of space. In order to substantiate this, one would now have to compare the results from G class clusters with clusters based on the spatial meanings of the relevant case markers. We leave this for future research. 7. Conclusions This study has looked into patterns of non-default case assignment as possible evidence for semantic role clustering. In order to do so, we have combined methods that allow for the detection of categories that are organized around prototypes as well as the detection of continua and cross-cluster similarities. While the dataset



Semantic role clustering

is limited, it does suggest that some of the traditionally assumed role clusters play a cross-linguistically significant role: experiencers, undergoers of body processes, and cognizers/perceivers in one- and two-place predicates; sources and transmitted speech/concept in three-place predicates. Our study provides no support for a categorial distinction between themes and instruments, but weak trends for a continuum between themes of transfer and attached items and, as a possibly distinct cluster, goals of transfer. This suggests that in these areas, languages treat predicate-specific roles in a much more varied way than traditional theories would lead us to expect. In the introductory sections, we contrasted our approach with alternatives that are based on pre-compiled predicate lists for which case frames can be surveyed and compared. Both approaches come with problems — our approach with the problem of noise in the data (individual predicates may be forgotten when non-default case frames are described in a language or compositional patterns may be misanalyzed as lexicalized complex predicates), list-based approaches with the problem of a priori skewing of results (specific predicate types may be overrepresented or underrepresented when drawing up a list and universal semantic analyses may be forced upon languages, artificially reducing diversity). Ultimately one will want to compare the results of both approaches. Hypotheses of universally relevant semantic role clusters need to pass muster across methods. Finally, any such hypothesis should be taken to be tentative before it is subjected to a diachronic and area-byarea analysis that probes into the dynamics of role clusters over time and space.

Acknowledgements Versions of this chapter were presented at the workshop on role complexes in Zurich, 4–5 April 2011, and the conference on non-canonically case-marked subjects in Reykjavik, 4–8 June 2012. We thank both audiences for helpful comments and questions. Author contributions: BB, AWM, and TZ conceived and designed the study. BB and TZ conducted the statistical analysis. LB and AWM did most of the data analysis and database coding. TZ did most of the data aggregation work. BB and AWM wrote the chapter. We thank Kevin Bätscher for help in data collection, Lukas Wiget for help in preparing the graphs, and Sergey Say for comments on an earlier draft. We also thank three anonymous reviewers and an anonymous editorial board member for helpful suggestions and requests for clarification.

References Arkadiev, Peter M. 2008. Differential argument marking in two-term case systems and its implications for the general theory of case marking. In Peter de Swart & Helen de Hoop (eds.), Differential subject marking, 151–171. Dordrecht: Springer.

75

76

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich Barddal, Johanna. 2008. Productivity: Evidence from case and argument structure in Icelandic. Amsterdam: Benjamins. Bickel, Balthasar. 2011. Grammatical relations typology. In Jae Jung Song (ed.), The Oxford handbook of language typology, 399–444. Oxford: Oxford University Press. Bickel, Balthasar & Johanna Nichols. 2009. Case marking and alignment. In Andrej Malchukov & Andrew Spencer (eds.), The Oxford handbook of case, 304–321. Oxford: Oxford University Press. Bickel, Balthasar, Manoj Rai, Netra Paudyal, Goma Banjade, Toya Nath Bhatta, Martin Gaenszle, Elena Lieven, Iccha Purna Rai, Novel K. Rai & Sabine Stoll. 2010. The syntax of three-argument verbs in Chintang and Belhare (Southeastern Kiranti). In Andrej Malchukov, Martin Haspelmath & Bernard Comrie (eds.), Studies in ditransitive constructions: A comparative handbook, 382–408. Berlin: Mouton de Gruyter. Bresnan, Joan & Jonni Kanerva. 1989. Locative inversion in Chichewa: A case study in factorization in grammar. Linguistic Inquiry 20. 1–50. Bryant, David & Vincent Moulton. 2004. Neighbor-Net: An agglomerative method for the construction of phylogenetic networks. Molecular Biology and Evolution 21. 255–265.  doi: 10.1093/molbev/msh018

Butt, Miriam. 2008. Case in Lexical-Functional Grammar. In Andrej Malchukov & Andrew Spencer (eds.), The Oxford handbook ofcase, 59–71. Oxford: Oxford University Press. Comrie, Bernard. 1978. Ergativity. In Winfred P. Lehmann (ed.), Syntactic typology: Studies in the phenomenology oflanguage. Austin: University of Texas Press. Comrie, Bernard. 1981. Language universals and linguistic typology. Chicago: University of Chicago Press. Comrie, Bernard. 2005. Alignment of case marking. In Martin Haspelmath, Matthew S. Dryer, David Gil & Bernard Comrie (eds.), The world atlas of language structures, 398–405. Oxford: Oxford University Press. Comrie, Bernard & Andrej Malchukov (eds.). 2015. Valency classes: A comparative handbook. Berlin: de Gruyter Mouton. Croft, William. 2001. Radical Construction Grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780198299554.001.0001 Croft, William. 2012. Verbs: Aspect and causal structure. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199248582.001.0001

Dalrymple, Mary & Irina Nikolaeva. 2011. Objects and information structure. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9780511993473 Dixon, R. M. W. 1994. Ergativity. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9780511611896

Dowty, David R. 1991. Thematic proto-roles and argument selection. Language 67. 547– 619.  doi: 10.1353/lan.1991.0021 Dryer, Matthew S. 1989. Large linguistic areas and language sampling. Studies in Language 13. 257–292.  doi: 10.1075/sl.13.2.03dry Dryer, Matthew S. 1997. Are grammatical relations universal? In Joan Bybee, John Haiman & Sandra A. Thompson (eds.), Essays on language function and language type dedicated to T. Givon, 117–143. Amsterdam: Benjamins. Haspelmath, Martin. 2001. Non-canonical marking of core arguments in European languages. In Aleksandra Y. Aikhenvald, R. M. W. Dixon & Masayuki Onishi (eds.), Non-canonical marking ofsubjects and objects, 53–84. Amsterdam: Benjamins.  doi: 10.1075/tsl.46.04has



Semantic role clustering

Haspelmath, Martin. 2011. On S, A, P, T, and R as comparative concepts for alignment typology. Linguistic Typology 15. 535–567.  doi: 10.1515/LITY.2011.035 Holisky, Dee A. 1987. The case of the intransitive subject in Tsova-Tush (Batsbi). Lingua 71. 103–132.  doi: 10.1016/0024-3841(87)90069-6 Huson, Daniel H. & David Bryant. 2006. Application of phylogenetic networks in evolutionary studies. Molecular Biology and Evolution 23. 254–267.  doi: 10.1093/molbev/msj030 Kaufman, Leonard & Peter J. Rousseeuw. 1990. Finding groups in data: An introduction to cluster analysis. New York: Wiley. Maechler, Martin, Peter J. Rousseeuw, Anja Struyf & Mia Hubert. 2005. cluster: Cluster analysis basics and extensions. R package, http://www.R-project.org/. Malchukov, A. 2005. Case pattern splits, verb types and construction competition. In Mengistu Ambember & Helen de Hoop (eds.), Competition and variation in natural languages: the case for case, 73–118. Amsterdam: Elsevier.  doi: 10.1016/B978-008044651-6/50006-9 Malchukov, Andrej, Martin Haspelmath & Bernard Comrie. 2010. Ditransitive constructions: A typological overview. In Studies in ditransitive constructions: a comparative handbook, 1–35. Berlin: de Gruyter Mouton. Merlan, Francesca. 1985. Split intransitivity: Functional oppositions in inflections. In Johanna Nichols & Anthony C. Woodbury (eds.), Grammar inside and outside the clause: Some approaches to theory from the field, 324–62. Cambridge: Cambridge University Press. Molochieva, Zarina. 2010. Tense, aspect, and mood in Chechen. Leipzig: University of Leipzig PhD dissertation. Nichols, Johanna. 2008. Why are stative-active languages rare in Eurasia? A typological perspective on split subject marking. In Mark Donohue & S0ren Wichmann (eds.), The typology ofsemantic alignment, 121–139. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199238385.003.0005

Nichols, Johanna, Alena Witzlack-Makarevich & Balthasar Bickel. 2013. The autotyp genealogy and geography database: 2013 release. Electronic database, http://www.uzh.ch/spw/autotyp. Onishi, Masayuki. 2001. Introduction: Non-canonically marked subjects and objects: Parameters and properties. In Aleksandra Y. Aikhenvald, R. M. W. Dixon & Masayuki Onishi (eds.), Non-canonical marking of subjects and objects, 1–52. Amsterdam: Benjamins.  doi: 10.1075/tsl.46.03oni

Pandharipande, Rajeshwari V. 1997. Marathi. London: Routledge. Primus, Beatrice. 1999. Cases and thematic roles. Tubingen: Niemeyer.  doi: 10.1515/9783110912463 Primus, Beatrice. 2006. Mismatches in semantic role hierarchies and the dimensions of role semantics. In Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie & Angela D. Friederici (eds.), Semantic role universals and argument linking: Theoretical, typological and psycholinguistic perspectives, 53–87. Berlin: Mouton de Gruyter. Pustet, Regina. 2002. Split intransitivity revisited: Comparing Lakota and Osage. International Jounral of American Linguistics 68(4). 381–427.  doi: 10.1086/466499 R Development Core Team. 2013. R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing, http://www.r-project.org. Rokach, Lior. 2010. A survey of clustering algorithms. In Oded Maimon & Lior Rokach (eds.), Data mining and knowledge discovery handbook, 269–298. New York: Springer [second edition]. Sava, Graziano. 2005. A grammar of Ts’amakko. Cologne: Köppe.

77

78

Balthasar Bickel, Taras Zakharko, Lennart Bierkandt and Alena Witzlack-Makarevich Say, Sergey. 2011. Nekanoničeskoe markirovanie aktantov mnogomestnyx predikatov: opyt kvantitativno-tipologičeskogo issledovanija. Acta Linguistica Petropolitana 7. 424–430 [http://iling.spb.ru/pdf/alp/alp_VII_3.pdf]. Schikowski, Robert. 2013. Object-conditioned differential marking in Chintang and Nepali. Zurich: University of Zurich PhD dissertation. Song, Jae Jung. 2001. Linguistic typology: Morphology and syntax. Harlow: Longman. Tsunoda, Tasaku. 1985. Remarks on Transitivity. Journal of Linguistics 21. 385–396.  doi: 10.1017/S0022226700010318

Tsunoda, Tasaku. 2004. Issues in case-marking. In Peri Bhaskararao & Karumuri Venkata Subbarao (eds.), Non-nominative subjects, vol. 2, 197–208. Amsterdam: Benjamins.  doi: 10.1075/tsl.61.11tsu

Van Valin, Robert D., Jr. 2005. Exploring the syntax-semantics interface. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9780511610578 Van Valin, Robert D., Jr. & David P. Wilkins. 1996. The case for ‘effector’: Case roles, agents, and agentivity revisited. In Masayoshi Shibatani & Sandra A. Thompson (eds.), Grammatical constructions, 289–322. Oxford: Oxford University Press. Witzlack-Makarevich, Alena. 2011. Typological variation in grammatical relations. Leipzig: University of Leipzig PhD dissertation.

Semantic roles and verbless constructions A Finnish challenge for verb-centered approaches Katja Västi and Seppo Kittilä

University of Oulu & University of Helsinki / University of Helsinki

Semantic roles constitute one of the most notorious notions in linguistics because they have been defined in numerous ways depending on the author’s theoretical framework and goals. Typically these definitions are somehow, more or less explicitly, based on verbs and their properties. In this chapter, semantic roles are discussed from a completely different perspective; we examine genuinely verbless constructions attested in Finnish newspaper headlines. The chapter addresses three main questions: First, what kinds of constructions do not need a finite verb to express dynamic events? Second, what kind of information remains unconceptualized in verbless constructions? And finally, what are semantic roles based on if there is no verb? The goal of the chapter is to show that verbs are not needed to define semantic roles, as an array of semantic roles can be recognized even in constructions lacking a verb. Keywords: semantic roles, verbless constructions, Cognitive Construction Grammar, Finnish, headlines

1. Introduction Semantic roles and closely related concepts, such as theta roles, thematic roles, and thematic relations, constitute a recurring topic in linguistic studies. They have been defined in a variety of ways depending on the various scholars and their purposes. However, as recently pointed out by Newmeyer (2010: 689), semantic roles represent one of the most notorious and ill-defined concepts in linguistics. Already nearly two decades before Newmeyer, Dowty (1991) discussed the range of definitions and functions in linguistic theories but, as repeatedly noted in the literature, there is still no consensus on, for example, which semantic roles exist, how they should be defined, what the nature of the roles is, and the correct set of roles we need for an adequate description of languages.

doi 10.1075/bct.88.04vas 2016 © John Benjamins Publishing Company

80 Katja Västi and Seppo Kittilä

In most cases, semantic roles have been taken as lexical properties of verbs; verbs are seen to assign semantic roles to their arguments. Despite a verb-centered approach, semantic roles themselves have often been defined independently of verbs as semantic primitives (e.g. Gruber 1965; Fillmore 1968, 1970; Jackendoff 1972, 1976) or as bundles of features (e.g. Rozwadowska 1988, 1989; Reinhart 1996, 2000, 2001, 2002). On the other hand, Dowty (1989, 1991) defines semantic roles in direct relation to verbs. He sees them as prototype categories that consist of recurring clusters of lexical entailments imposed by groups of verbs on their arguments. Furthermore, Næss (2003, 2007) introduces bundles of features that resemble those in the work of Rozwadowska (1988, 1989) but do not directly correspond to semantic roles seen as subcategorized by verbs. Instead, she intends that terms like agent and patient indicate “labels for clusters of properties exhibited by noun phrases (or, strictly speaking, by their referents) when these function as core arguments of specific clauses” (Næss 2003: 106, 2007: 37). The later work of both Fillmore and his colleagues on Frame Semantics and the FrameNet Project encompasses elaborate notions of several linguistically relevant role types, of which “frame elements” and “semantic roles” relate most directly to the aforementioned notions: frame elements are event participants involved in semantic frames evoked by lexical units, whereas semantic roles are their “linguistically motivated abstractions in that they pick out specifically those properties that tend to display the same behavior in morphosyntax” (Fried & Östman 2004: 42). Finally, in her Cognitive Construction Grammar approach, Goldberg (1995, 2006) allows for constructions to assign arguments “argument roles”, while verbs are associated with “participant roles”. This division partly resembles the difference between frame elements and semantic roles in Fillmorean Construction Grammar, but the role types are given different theoretical statuses in the two constructional frameworks. In addition to numerous definitions of semantic roles, the number of roles distinguished also varies drastically. For example, when dealing with subject and object selection, theories of so-called generalized semantic roles distinguish only two roles; in Role and Reference Grammar they are labeled as actor and undergoer. These “macroroles” subsume several more specific thematic relations that are based on verbal semantics (Van Valin & LaPolla 1997: 139–142). In Basic Linguistic Theory (see e.g. Dixon 2010), the number of semantic roles is usually around 15–20; in addition to agents and patients, roles such as instrument, recipient, beneficiary, path, comitative, source, and location are acknowledged. As a rule of thumb, fewer roles are needed for syntax than for lexical semantics. Regardless of the number of roles distinguished, what most current accounts of semantic roles have in common is that they employ, in one way or another, and more or less explicitly, events and their properties when defining the semantic



Semantic roles and verbless constructions

roles of linguistic elements. For example, Croft (1991: 149–182) suggests that the causal structure of events provides relevant semantic features on which semantic roles are based. He analyzes events by verbal decomposition, which takes the form of a causal chain. DeLancey (1991), for his part, presents a localist theory of event structure for characterizing semantic roles. Næss (2003, 2007) also defines semantic roles as participants in events, even though no direct reference to verbs is made. In other words, the events coded constitute the starting point when semantic roles are examined, and as they are coded by verbs, verbs are central in most definitions, although it is important to state clearly that not all scholars use verbs explicitly for this purpose. The present paper adds an entry to the list of studies of semantic roles, but the roles are studied from a drastically different viewpoint. We examine semantic roles in constructions that lack a verb altogether, which naturally means that verbal semantics cannot be responsible for the role assignment. Similarly to Næss (2003: 104–106, 2007: 35–37), we define semantic roles as event-based semantic generalizations without assuming that they are subcategorized by verbs. In addition, our approach shares features with the Cognitive Construction Grammar of Goldberg (1995, 2006), where two types of roles are identified: more coarsegrained semantic roles, or argument roles, are associated directly with argument structure constructions, and fairly detailed participant roles, for their part, are associated with semantic frames, which may be expressed with a number of different verbs. In instances of argument structure constructions, these roles are fused in principled ways (see Goldberg 1995: 43–66). However, as we are discussing nonelliptic verbless constructions, their participant roles cannot be defined in relation to any verb either. Instead, we assume that participant roles are associated directly with semantic frames evoked by lexical units used in constructions. In fact, Goldberg (2006: 8) also took notice of verbless constructions: “Many languages have constructions in which no verb is expressed at all. These cases are prime examples of argument structure constructions, since their meaning cannot naturally be attributed to a (non-existent) verb.” She does not address the issue in more detail, but we will elaborate on these constructions by discussing their inherent meanings including the semantic roles of their arguments. Our paper addresses three main topics. First, we will look at the kinds of constructions that do not need a finite verb to express dynamic events. Second, we are interested in the kind of information on events that remains unconceptualized if a given construction does not involve a verb. Finally, on the basis of the previous questions, we will investigate what semantic roles are based on in verbless constructions. All of this will be done by analyzing how the nominal elements in verbless constructions are formed and organized in order to express events independently, and how their lexical semantics, combined with different pragmatic

81

82

Katja Västi and Seppo Kittilä

factors such as world knowledge, contributes to their readings. The discussion in this chapter is based exclusively on Finnish, but due to the strong theoretical flavor of the discussion, we believe that the findings of this chapter have greater general relevance for our understanding of semantic roles in other languages as well. The organization of this chapter is as follows. In Section 2, we will discuss previous accounts of semantic roles in more detail. This is necessary for the discussion in Sections 4 and 5. In Section 3, we present our data. In Section 4, the three questions noted above will be examined, and Section 5 summarizes the most important findings of the chapter. 2. Previous approaches to semantic roles In what follows, we will discuss some of the basic ways in which semantic roles have been defined in previous studies. In other words, our goal is to briefly discuss what has been done thus far regarding semantic roles; our own way of defining semantic roles will be elaborated in Section 3. It is not the purpose of this section to exhaust the field by discussing all possible definitions of semantic roles, but we will focus on definitions that are relevant to the discussion in this chapter. A typical, while in some cases only implicit, way to see semantic roles is that verbs constitute the basis by referring to the event denoted, while arguments are responsible for specifying the identity and the nature of participants present in typical clauses (see Eriksen et al. 2010 for a discussion of this). In this view, the verb ‘eat’ in ‘the child is eating vegetables’ refers to the given event, while the nouns ‘child’ and ‘vegetables’ specify who is doing what to whom/what. Put another way, the verb assigns semantic roles to the arguments, and, depending on the level of abstraction, we may call ‘child’, for instance, the (affected) agent or eater and ‘vegetables’ the patient or the thing eaten. The verb that refers to the event in question is thus responsible for the semantic roles. On the other hand, the roles borne by participants may vary with different verbs. For example, one and the same participant may be more or less agentive, or more or less affected depending on context. It is important to note that individual roles are not dependent on verbs and that verbs do not assign semantic roles to arguments in a dichotomous manner, but the degree of agency or affectedness may naturally vary. For example, the causer — understood in a broad sense as covering all causers/causes of events, not only canonical agents — of ‘break’ can be a canonical agent, an involuntary agent, or a force. However, it has been argued by, for example, Eriksen et al. (2010) that there are cases in which the division of labor between arguments and verbs may deviate from the aforementioned principle, and that there are events which cannot be divided into meaningful components. Illustrative examples are represented



Semantic roles and verbless constructions

by meteorological events. These cannot be divided into more and less active participants in the same manner as typical events, such as ‘the child is eating fruit’, can (see also Croft 1991: 141ff). They also lack salient human participants, which makes them different from the canonical events. This has the consequence that meteorological events can, in principle, be exhaustively described by a predicate or an argument alone. For example in (1), the verb ‘come’ accompanying the noun is semantically rather vacuous and has a grammatical function: (1)

Korean (Jae Jung Song, p.c.) Pi-ka /nwun-i /wupak-i o-nta. rain-nom /snow-nom /hail-nom come-ind ‘It is raining/snowing/hailing.’

Second, as was already hinted at above and as has been discussed by, for example, DeLancey (1984) and Dowty (1991) for agent, and Rappaport Hovav & Levin (2008) for goal and recipient, verbs do not strictly assign their arguments a single role but allow variation. One of the roles, such as Dowty’s Proto-Agent, may be seen as the expected role, while deviations from it need to be highlighted, for example, by modifying the marking of argument(s). Consider: Lezgian (Haspelmath 1993: 292) (2) a. zamira-di get’e xa-na Zamira-erg pot break-aor ‘Zamira broke the pot.’ b. zamira.di-waj get’e xa-na Zamira-adel pot break-aor ‘Zamira broke the pot accidentally/involuntarily.’ c. * didedi-waj gam xkaž-na mother-adel rug lift-aor For: ‘Mother accidentally lifted the rug.’

In Lezgian, the verb ‘break’, along with other labile verbs, allows two roles, which can be labeled as agent and involuntary agent. In (2a), A (i.e. the agent argument) appears in the ergative case and refers to a canonical agent. Example (2b) illustrates the Involuntary Agent Construction of Lezgian; A appears in the adelative case, and the referent of A causes the event involuntarily. (2b) illustrates a typical example, where the deviant reading is a combination of case marking and verb semantics. A labile verb combined with adelative coding results in an involuntary reading. It should also be noted that the adelative coding of A is not possible with ordinary transitive verbs, such as ‘lift’, which makes (2c) ungrammatical (see Kittilä 2005 for a more detailed discussion). This underlines the fact that non-canonical marking is more intimately associated with a certain semantic role. In the canonical cases,

83

84

Katja Västi and Seppo Kittilä

such as (2a), in turn, the arguments coded by grammatical cases (ergative and absolutive in Lezgian) are more directly related to the semantics of the verb. Third, the lexical semantics of arguments has been shown to play a role in determining their semantic roles (see e.g. Rappaport Hovav & Levin 2008; Kittilä 2008 for the effects of animacy on goals and recipients; Song 2011 for agents). The roles borne by arguments may be sensitive to their own semantic nature with certain verbs. Examples include cases such as ‘John sent the parcel to Lisa’ and ‘John sent the parcel to Dunedin’. In the first case, the event denoted may, for example, be said to include an agent (or source), theme, and a recipient, while in the second case, we have a goal instead of a recipient. The main difference lies in the animacy of R (see Kittilä 2008 who also admits that animacy is not the only difference between the arguments, but their semantic roles also vary). Differently from (2), the lexical semantics of the given arguments is responsible for the attested differences. In (2a–b), the referent of A is potentially a canonical agent due to animacy, and (2a) and (2b) differ in that only in (2a) does the agent act volitionally. In the sending events discussed, in turn, only animate endpoints of transfer have the potential of being recipients. Finally, in Cognitive Construction Grammar (Goldberg 1995, 2006), semantic roles may be seen as constructional properties in such a way that the lexical semantic features of a noun do not strictly define the roles that it can bear, nor do lexical semantic features of a verb directly determine the roles that can appear in a clause together with it. Instead, semantic roles are based on constructions that encode event types basic to human experience (Goldberg 1995: 39–40). Speakers may construe states of affairs in the world from different vantage points and thus portray individual participants in different roles (see also DeLancey 1991). Typically, it is possible to use a single verb in different constructions, which then impose different semantic roles as well as other semantic and pragmatic properties on their arguments. For instance, the semantic role of a golden key has traditionally been labeled as instrument in both Hannah opened the gate with a golden key and A golden key opened the gate, whereas from the perspective of Cognitive Construction Grammar it is possible to analyze the latter case such that a golden key is also conceptualized metaphorically as an agent (or force, if the need arises to differentiate between animate and inanimate instigators). In other words, cognitive linguistics aspires to analyze how the world can be construed using language, not the world itself, and to recognize conventional pairings of form and meaning (e.g. Leino 2001: 34). Nevertheless, this does not exclude the possibility of analyzing verbal semantics at a different level of description, that is, in terms of participant roles, which are extralinguistic entities. In Cognitive Construction Grammar this premise can be seen in the formulation that argument structure constructions bear semantic roles that are semantically constrained relational slots in the



Semantic roles and verbless constructions

dynamic scene associated with the construction. They are fused with participant roles, which, for their part, are relational slots in the dynamic scene associated with the verb (Goldberg 1995: 43–52). In addition to the different theoretical statuses given to semantic roles, there are clear differences in how the differences between suggested roles are manifested and how seriously formal differences between the roles should be or have been taken. When should we speak of distinct roles and when not? In the most evident cases, formal differences are semantically motivated, as in (3) (see also (2)): Finnish (p.k.) (3) a. lapsi rikko-i maljako-n child break-3sg.pst vase-acc ‘A/the child broke a/the vase.’ b. lapsi vihaa maljakko-a child hate.3sg.prs vase-ptv ‘A/the child hates a/the vase.’

The event denoted in (3a) can be said to include an agent and a patient, while in (3b) the roles present can be defined as experiencer and stimulus. This difference is also formally manifest in that O (i.e. the object of a transitive clause) bears accusative coding in (3a), while in (3b) it appears in the partitive case. In (2) from Lezgian, the patient role is maintained but there are evident differences in the agent role. In (2a), we are dealing with a canonical agent, while in (2b), the event denoted is instigated accidentally and the agent is best labeled as an involuntary agent in the sense of Haspelmath (1993: 292) and Kittilä (2005). In (4) and (5), in turn, the differences are merely semantic: Finnish (p.k.) (4) a. lapsi maalas-i maljako-n child paint-3sg.pst vase-acc ‘A/the child painted a/the vase.’ b. lapsi näk-i maljako-n child see-3sg.pst vase-acc ‘A/the child saw a/the vase.’ Lezgian (Haspelmath 1993: 292; Moor 1985: 112) (5) a. zamira-di get’e xa-na Zamira-erg pot break-aor ‘Zamira broke the pot.’ b. ničaγval-di ruš q’ena disease-erg girl killed ‘The disease killed the girl.’

85

86 Katja Västi and Seppo Kittilä

Similarly to (2)–(3), examples (4a)–(5a) can be analyzed as including an agent and a patient, and the marking pattern is nom-acc or abs-erg, depending on the language. As in (3b), we have a similar event in (4b) with an experiencer and a stimulus. However, in contrast to (3b), the semantic difference is not manifested formally. In (5), we are dealing with a similar case; differences between canonical agents and forces are not made formally explicit. Semantically, the differences between the roles are evident as they are in (2) and (3), but they have no formal realization. Semantic roles are descriptive and theoretical tools developed and used by linguists to analyze and describe languages. This makes formal differences central to their definition, especially in grammatical description; in verbal semantics, in turn, formal differences are less relevant. In other words, any well-grounded distinction between semantic roles should be based on formal evidence (see e.g. Lazard 1998: 163–166; Næss 2007: Chapter 3). This makes examples (4) and (5) potentially problematic for the analysis of semantic roles. (However, a considerable part of the problem is that, in individual studies, particular roles are often not defined in any particular way, so it is not at all clear whether certain labels of semantic roles refer to the same roles or not in different studies.) Put another way, we do have a distinction between patients and stimuli (of emotions) in Finnish, and between prototypical and involuntary agents in Lezgian, because these roles are formally distinct. On the other hand, we do not have a distinction between forces and agents in Lezgian, and patients and other stimuli in Finnish. However, differences between semantic roles cannot be made in such a dichotomous manner, but other things must be considered. A case in point is illustrated in (6): Finnish (p.k.) (6) a. opettaja lähett-i kirja-n lapse-lle teacher send-3sg.pst book-acc child-all ‘The teacher sent a/the book to the child.’ (recipient) b. opettaja lähett-i lapse-n talo-lle teacher send-3sg.pst child-acc house-all ‘The teacher sent a/the child to the house.’ (goal) c. opettaja lähett-i lapse-n rehtori-n luo teacher send-3sg.pst child-acc principal-gen to ‘The teacher sent a/the child to the principal.’ (vicinal goal)

The examples in (6) illustrate three roles labeled as recipient (6a), goal (6b), and vicinal goal (6c) by Kittilä & Ylikoski (2011). Semantically, as the authors discuss in detail, goal and vicinal goal are close to each other as they both denote endpoints of motion without caused possession, and they are clearly different from the recipient, which involves caused possession as well. Despite this, goal and recipient



Semantic roles and verbless constructions

are accorded the same formal treatment and marked differently from the vicinal goal in Finnish. If we then rely on formal evidence for defining semantic roles, the most natural conclusion is that we have two roles, a goal/recipient role and a vicinal goal role. However, this kind of definition does not seem satisfactory, as the semantically close roles are seen as distinct. The examples in (6) thus show that formal evidence does not always provide us with a definitive answer. In (6), the differences and similarities in coding follow different principles. The roles of goal and recipient are sufficiently distinguished from each other based on animacy; goals are inanimate, while recipients are animate. This renders a formal distinction superfluous. On the other hand, the role of vicinal goal is not directly retrievable from animacy or any other feature. Moreover, vicinal goals resemble goals, but the two roles are nevertheless distinct, which makes it necessary to resort to formal means to distinguish them explicitly. The three goal roles (understood in a broad sense) show that arguments are not marked solely based on their semantic roles but other factors also make a contribution, which goes further to make defining semantic roles a tricky issue. 3. Data We have based our study on 500 instances of verbless constructions used as newspaper headlines collected from the Finnish Language Bank.1 More precisely, the data was collected from the Finnish Text Collection, a selection of electronic research material containing written Finnish beginning in the 1990s. There are several subcorpora, among which Aamulehti 1999 was chosen. Aamulehti is a daily newspaper published in Tampere (in southern Finland), and the number indicates its year of publication. We decided to employ newspaper language data, because it is considerably easier to retrieve clause-like verbless expressions from written language corpora than from spoken language corpora, and because verbless constructions are particularly common in headlines.2 This is not, however, to say that verbless constructions do not occur in spoken language (see e.g. Helasvuo 2001: 105–131 and references therein). In order to illustrate certain points better, 1.  For more information on the Language Bank, see . 2.  Despite the fact that our data consists of headlines, our purpose is not to discuss grammatical, textual, or discourse properties specific to headlines. There is quite a long, albeit not very extensive, tradition of studying grammar as well as other features of headlines in different languages (see e.g. Straumann 1935; Åkermalm 1965; Conon 1973; Mårdh 1980; Simon-Vandenbergen 1981; Vahtera 2009), but they cover aspects different from ours (cf., however, Schneider 2000).

87

88

Katja Västi and Seppo Kittilä

we present elicited examples as well. It should also be noted that we used the corpus to chart the possible types of verbless constructions that are highly clause-like and independent and which express events, but we have not analyzed the data in any statistical way. When speaking of constructions, we mean, following Goldberg (2006: 5), “learned pairings of form with semantic or discourse function”. She details the definition of a construction as follows: “Any linguistic pattern is recognized as a construction as long as some aspect of its form or function is not strictly predictable from its component parts or from other constructions recognized to exist. In addition, patterns are stored as constructions even if they are fully predictable as long as they occur with sufficient frequency.” There are many kinds of verbless constructions, such as predicate nominal and possessive constructions. As the present study concerns semantic roles, we focus on those genuinely verbless constructions that do not result from ellipsis, that can be construed as indicating dynamic events (instead of states), and that can be seen as productive or at least semi-productive. That is, our work goes far beyond studies of missing copulas and other similar constructions. Naturally, such headline data varies structurally. What the investigated constructions have in common, however, is that they all include a noun phrase in a grammatical case (NPgram), namely the nominative or partitive. (The third grammatical case of Finnish, the accusative, is not attested in our data.) In addition, most of the investigated constructions include an oblique phrase (OblP) which may be another noun phrase in a so-called local case3 (NPloc, cf. (7a)), an adposition phrase (cf. (8)), or an adverb phrase (cf. (9a–b)). Thus, at a minimum, the investigated constructions consist of two arguments, but a third argument — another oblique phrase — is possible as well (cf. (7b)). Adjuncts, such as temporal modifiers, are also common (see (9b)). Finnish (Language Bank) (7) a. Suomalaislaittee-t kuu-hun Finnish.appliance-nom.pl moon-ill e.g. ‘Finnish appliances go / are sent to the moon.’4

3.  These are the inessive (‘in’), elative (‘from (inside)’), illative (‘(in)to’), adessive (‘on, at, by, near’), ablative (‘from’), allative (‘(on)to, for’), essive (‘as’), or translative (‘(changing) into’). For Finnish local cases, see e.g. Huumo & Ojutkangas (2006) and references therein. Note that many Finnish adverbs also contain a local case suffix. 4.  Note that it is a purely practical decision to translate instances of verbless constructions with a particular mood and tense. Construing them as denoting, for instance, the past is connected with pragmatics; the construction in itself does not involve any information on mood or tense.

Semantic roles and verbless constructions 89



b.

Fortumi-lle kilpailija länne-stä? Fortum-all rival west-ela [company name] e.g. ‘Fortum gets a rival from the west?’

(8) Outo-a valo-a Tanska-n ja Ruotsi-n yllä strange-ptv light-ptv Denmark-gen and Sweden-gen above e.g. ‘A strange light was seen above Denmark and Sweden.’ (9) a. b.

Väestötietojärjestelmä alas population.register.system down e.g. ‘The population register system went down.’ Kunna-n vero-t entisellään ensi vuon-na municipality-gen tax-nom.pl unchanged next year-ess e.g. ‘The taxes of the municipality are kept unchanged next year.’

Two instances in our data contain a nonfinite verb form expressing manner or means (InfP, cf. (10)) occupying the slot of the oblique phrase. In three instances, the construction consists of an NPptv and a quantifying expression (cf. (11)). Furthermore, there are three instances of a relatively idiomatic construction consisting of an NPnom and an expression of age (cf. (12)). (10)

Finnish (Language Bank) Uusi vuosi luistel-len new year skate-inf e.g. ‘The new year goes skating.’

(11) Huume-i-ta entis-tä enemmän drug-pl-ptv former-ptv more e.g. ‘Drugs are used even more than before.’ (12) Kaupi-n sairaala 60 vuot-ta Kauppi-gen hospital 60 year-ptv e.g. ‘The Kauppi hospital celebrated 60 years of existence.’

In the next section, we will provide a more detailed description of the abovementioned constructions in the light of analyzing what dynamic construals of verbless expressions are based on. However, the focus of our examination will not be on specific morphosyntactic characteristics of Finnish verbless constructions, as we believe that our approach could be applied to other languages as well.

90 Katja Västi and Seppo Kittilä

4. Semantic roles and verbless constructions 4.1 Dynamicity without verbs In this section, the data presented in Section 3 is discussed from the perspective of how verbless constructions are able to express dynamic events without a verb. We have not therefore attempted to exhaust all the possible verbless constructions that could fit into the scope of the present study. Instead, we intend our analysis, emerging from our data, to be an initial contribution to this topic. We see semantic roles as event-based bundles of semantic features that are constructional attributes and thus connected with certain morphosyntactic features (cf. argument roles in the spirit of Goldberg 1995, 2006). It is therefore possible to draw a parallel between basic sentence types that involve a verb and the verbless constructions investigated in the present study. On these grounds, our approach includes adopting Goldberg’s Scene Encoding Hypothesis that states “constructions which correspond to basic sentence types encode as their central senses event types that are basic to human experience”. Event types, for their part, are relatively abstract classifications of different states of affairs in the world. (Goldberg 1995: 39–40.) Semantic roles thus indicate participants conceptualized as conventional parts of some basic event type. For defining semantic roles, this means that it is crucial to recognize the event type denoted by the construction in question, and if verbless constructions encode event types, they necessarily involve semantic roles. Events are indisputably denoted by verbs. However, in many languages, static events, or rather states, such as existence or possession, can be expressed without (copula) verbs (see Payne 1997: 113ff for a discussion). However, our data clearly shows that some dynamic event types can also be expressed without verbs (see Västi 2011a, 2011b, 2012 for senses of two specific constructions; see also Vahtera 2009 for a comparison between Swedish and Finnish, and Kopotev 2007a, 2007b for a comparison between Russian and Finnish). Dynamic events are defined here as involving concrete motion or some kind of abstract change, which is typically coded by verbs. Thus, in verbless constructions, some element other than a verb must code the change or motion. Moreover, this element must have a relatively specific meaning which constrains possible verbless constructions. In constructions expressing highly transitive events with an explicit agent and patient, for example, arguments are typically marked with elements bearing extremely schematic meaning (such as the nominative and accusative case) if they are marked at all, which makes transitive events bad candidates for being expressed by verbless constructions. Nevertheless, as several examples above in Section 3 suggest and as will be seen later in this section, verbless constructions can be used to express even transitive events.



Semantic roles and verbless constructions

The core of most discussed cases is composed of an NP in a grammatical case and an OblP (see Section 3). These constructions are illustrated schematically in Table 1. Table 1 illustrates the possible construction types from the perspective of the relative word order of the first two arguments, which dictates whether the construction in question is existential or non-existential. In Finnish, linearity plays a role in differentiating between basic sentence types, and it contributes to the semantics of the constructions under investigation. A more thorough discussion of linearity, however, lies outside the scope of this chapter. We have labeled the constructions with the initial-position NPgram as non-existential, and the constructions with inverse word order as existential.5 This is in line with the tradition in Finnish linguistics and is applicable to the constructions studied in this chapter. The constructions are further divided into setting, source, and target constructions according to the meaning of the OblP, which may indicate the setting of the denoted event (‘event setting’), or either the source state (‘change from’) or target state (‘change to’) of the NPgram’s referent. The meanings of these constructions cover several event types in the spatial, possessive, temporal, identificational, and circumstantial domains, to name only the most typical ones.6 The ability of a verbless construction to express dynamic events is largely due to the OblP accompanying the NPgram. As for Finnish verbless constructions, Helasvuo (1991: 40–41, 2001: 123–125) and Västi (2011b: 46–50, 2012) have suggested that local case suffixes may function as their independent predicates, and the same analysis can be extended to apply to adpositions and adverbs. Kopotev (2007b: 129–131) has also proposed a somewhat similar theory of the existence of verbless constructions in Finnish. The approaches of both Helasvuo and Västi Table 1.  Typical verbless construction types denoting dynamic events in Finnish non-existential constructions

existential constructions

setting constructions [NPgram + ‘event setting’]

setting constructions [‘event setting’ + NPgram]

source constructions [NPgram + ‘change from’]

source constructions [‘change from’ + NPgram]

target constructions [NPgram + ‘change to’]

target constructions [‘change to’ + NPgram]

5.  For semantics of Finnish existential sentences, see e.g. Huumo (2003). 6.  Nevertheless, meanings of verbless existential constructions belonging to the circumstantial domain seem rather odd. There are no instances of those meanings in our data, nor are we able to intuitively form completely natural examples (?Lakko-on posti-n työntekijö-i-tä [strike-ill postal.service-gen employee-pl-ptv] e.g. ‘Postal workers go on strike.’).

91

92

Katja Västi and Seppo Kittilä

are founded on Siro (1964: 26–29) who introduced the term “quasi-predicate” to Finnish linguistics. The term quasi-predicate designates those local case suffixes and adpositions — and adverbs, we may add — that are used to mark the arguments of a verb. It “refers to” the subject in an intransitive clause and to the object in a transitive clause. This analysis can be reformulated so that, in an intransitive clause, the subject acts as an argument of the quasi-predicate and, respectively, the object acts as an argument of the quasi-predicate in a transitive clause. Thus, the quasi-predicate indicates a semantic relationship between its stem and argument. As regards verbless constructions, consider (13) which consists of an NP in the nominative case (Hägglund [surname]) and an NP in the illative case (Naton kokoukseen ‘to a NATO meeting’): (13)

Finnish (Language Bank) Hägglund Nato-n kokoukse-en Hägglund NATO-gen meeting-ill [surname] lit. ‘Hägglund to a/the NATO meeting.’

In (13), there is only one argument (Hägglund) in addition to the quasi-predicate’s (-en ‘to’) stem (Naton kokoukse- ‘NATO meeting’), and hence the expression involves the meaning ‘Hägglund to a/the NATO meeting’. In the absence of a verb, the illative case suffix can be analyzed as a proper predicate instead of a quasipredicate. In other words, in finite constructions, verbs act as predicates, and case suffixes, adpositions, and adverbs act as quasi-predicates, whereas in verbless constructions, case suffixes, adpositions, and adverbs are the only or the main predicating elements. However, we are not saying that verbs and case suffixes etc. would have equal statuses as predicates in any context, but that verbs predicate in much greater detail (see Section 4.2). At this point, it is important to note that Finnish verbless constructions tend to be highly polysemous (which probably applies to similar constructions in other languages as well). Polysemy is a natural attribute of constructions (see e.g. Goldberg 1995: 31–39), but it is particularly substantive in verbless constructions, as their individual instances can often be construed as encoding remarkably dissimilar event types (see Västi 2011b for a more detailed discussion of polysemy in verbless constructions, and Västi 2011a, 2012 for empirical support for our semantic analysis of verbless constructions). Probably the most central aspect behind the semantic variation is that, in most cases, the NPgram is construable as both a subjectlike and an object-like argument, and there is often even a third possibility where the construction receives an existential meaning. This directly affects the makeup of the event type. In (13) above, the first two contradicting construals manifest themselves as different perspectives on the denoted event. Should Hägglund be



Semantic roles and verbless constructions

interpreted as a subject-like argument, the event in question is understood as happening on Hägglund’s initiative; he goes to the NATO meeting voluntarily. On the other hand, if Hägglund is interpreted as an object-like argument, the event in question is understood as a transitive event without an explicitly mentioned actor; somebody causes Hägglund to go to the NATO meeting. That is, even though there is only one explicit argument, transitive construals are not excluded. Verbless constructions can also be construed as indicating dynamic events in the absence of an OblP with an inherently dynamic meaning, even if those construals are dynamic to a lesser degree compared with the example (13) discussed above. Consider (14), where the OblP is in a static case, namely the inessive: Finnish (Language Bank) (14) a. Verilöyly Honolulu-ssa massacre Honolulu-ine lit. ‘Massacre in Honolulu.’ b. Laaja puhdistus Albania-ssa=kin extensive cleansing Albania-ade=too lit. ‘Extensive cleansing in Albania too.’

In (14a), the inessive case functions as a predicate and verilöyly ‘massacre’ as its argument. The literal meaning of the expression is ‘A massacre in Honolulu’, which is not dynamic in itself. The lexical meaning of verilöyly, however, includes dynamicity because ‘massacre’ is an event, and should verilöyly be construed as a subjectlike argument, the expression approximates the meaning ‘A massacre took place in Honolulu’. Contrary to the intransitive construal of (13) above, this conceptualization does not contain an instigator, but the event is portrayed as spontaneous. On the other hand, should verilöyly be construed as object-like, the expression roughly means ‘A massacre was carried out in Honolulu’.7 A lexical meaning of an NPgram gains a particularly large role in dynamic construals of seemingly stative expressions in cases where the NPgram is deverbal (cf. (14b)): a deverbal NPgram specifies the nature of the process directly, which is typically not possible for verbless constructions (see Section 4.2). In any case, the pervasive ambiguity between event types applies to construing (14b) as well. That is, it may be interpreted as ‘There occurs an extensive cleansing in Albania too’ or as ‘An extensive cleansing is organized in Albania too’.

7.  One might think that the most neutral construal for a construction like (14a) would be that there simply is an unexpressed copula verb. However, the expression cannot mean ‘A massacre is in Honolulu’. Instead, it is possible to conceive Honolulussa as a postmodifier to the head noun verilöyly. In that case, Verilöyly Honolulussa is interpreted as an NP and it does not fall into the scope of the present study.

93

94 Katja Västi and Seppo Kittilä

In addition to the construction types discussed above, our data contains sporadic instances of the more fixed constructions exemplified in (10)–(12) in Section 3. They cannot be analyzed similarly to (13)–(14) above, i.e. as involving a predicating OblP that contributes to the dynamic meaning. Nevertheless, they too express events, so their dynamic meaning must be associated with a linguistic factor. In our data, there are only two instances of a type (10) construction that consists of an initial-position NPnom indicating time and of an InfP that indicates manner or means (but not the whole event). Consider (15) (see also (10) above): (15)

Finnish (Language Bank) Lomapäivä vaikka tanssi-en holiday for.instance dance-inf lit. ‘A/the holiday, for instance, dancing.’

The core elements of (15) are the NP lomapäivä ‘holiday’ and the InfP tanssien ‘dancing’ that together build a temporal profile for the expression. First, lomapäivä lexically refers to a span of time. Second, as a verb-form expression of manner, tanssien modifies an unspecified process which concerns the expressed holiday. In other words, tanssien suggests that a process indeed takes place. As in the previous cases (13)–(14), the NPgram lomapäivä may be construed as subject-like or objectlike, and thus the meaning of the expression may be glossed as ‘A/the holiday goes, for instance, dancing’ or as ‘A/the holiday is spent, for instance, dancing’. Moreover, our data contains three instances of a type (11) construction exemplified here with (16): (16)

Finnish (Language Bank) Kotikiel-tä liian vähän home.language-ptv too little lit. ‘Home language too little.’

The construction consists of an NPptv (here kotikieltä ‘home language’) and of a quantifying expression (here liian vähän ‘too little’), and it is especially the NPptv that seems to enable dynamic construals. Namely, the Finnish partitive is used to indicate, among other things, aspectual unboundedness, due to which the NPptv and the quantifying expression together have the potential of denoting a process. This unspecified, quantified action is, then, directed at the referent of the NPptv. The fundamental polysemy of verbless constructions can be seen here in alternative construals of the type ‘There is too little home language’ and ‘Home language is used too little’. Finally, a type (12) construction is also represented by three instances in our data. It is a relatively idiomatic construction, which consists of an NPnom and an expression of age in the form of years:



(17)

Semantic roles and verbless constructions

Finnish (Language Bank) Lielahde-n SPR 30 vuot-ta Lielahti-gen SPR 30 year-ptv [district] [The Finnish Red Cross] lit. ‘The SPR of Lielahti 30 years.’

Unlike the constructions discussed so far, this type only allows for one dynamic interpretation, namely ‘NPnom turns X years’, which is probably due to the construction’s idiomaticity and motivated by a semantic frame concerning birthdays, and not so much the construction’s grammatical and lexical elements as such. As regards semantic roles, this construction is not as illustrative as the ones introduced above, because the idiomaticity reduces the effect of the semantic principles associated with the other constructions. The discussion above is definitely not an exhaustive description of the semantic scope and variation in our data, but it should suffice for showing that, in verbless constructions, dynamicity can be associated with several different factors of which oblique elements with an inherently dynamic meaning are the most important ones. Several properties of individual elements influence particular construals, but they are not discussed in detail in this chapter. This is thus a rather preliminary analysis of linguistic elements contributing to the dynamic meanings of the verbless constructions investigated in the present paper. Our objective is to underline the important point that verbless constructions may also denote diverse events, as this justifies assigning semantic roles to their core building blocks. That is, semantic roles represent conceptualizations of typical participants in typical events. It is rather easy to find potentially comparable constructions in other languages, as in Swedish (e.g. Jeltsin till USA i september [Yeltsin to the USA in September]; see Vahtera 2009 for a description of Swedish verbless constructions), and it is our hope that several different languages will be considered explicitly in later studies. 4.2 Events coded by verbless constructions In the previous section, we showed that a verb is not necessary to express dynamic events, and analyzed the means by which verbless constructions may predicate. However, we may expect that conceptualizing an event with a verbless construction is not the same as conceptualizing that event with a full clause containing a finite verb, as verbs typically provide detailed information about the qualities of the event. We will therefore discuss basic characteristics of verbless conceptualizations next. Events are coded linguistically by argument structure constructions that include a reference to salient participants and relations between them. A typical division of labor between the elements of a construction is that a finite verb indicates

95

96 Katja Västi and Seppo Kittilä

the event while arguments specify its participants. For example, in Paula painted her nails, the verb ‘paint’ describes the event and the nouns identify its participants. However, as shown in the previous section, the overall picture is not quite as simple. In a language like Finnish, arguments also provide information about the denoted event; their coding reveals relations between participants, and those relations convey information about processes. This kind of information mainly concerns the given event type. That is, by means of arguments alone it is possible to express, for instance, causal relations and the direction of energy flow between participants, but not to explicate the details of an event. Consider the example in (18), consisting of two arguments: (18)

Finnish (Language Bank) Koulutukse-en lisä-ä raha-a education-ill more-ptv money-ptv lit. ‘To education more money.’

The illative suffix -en ‘(in)to’ functions as the predicate of the construction by specifying the relationship between its stem (koulutukse- ‘education’) and the other argument (lisää rahaa ‘more money’), much in the same way a verb would do this if present. Based on the inherent meaning of the illative, we can infer that more money goes to education, but we do not know exactly how. (18) can be interpreted from at least the two different perspectives presented in the previous section. That is, the NPgram lisää rahaa may be construed as a subject-like or an object-like argument, but manner, means, and comparable semantic attributes are not included in the expression’s profile. In contrast, the full clause in (19) exemplifies the way a verb specifies the nature of the process: (19)

Finnish (p.k.) Koulutukse-en kerä-tään/myönne-tään lisä-ä raha-a education-ill collect/grant-pass.prs more-ptv money-ptv ‘More money is collected/granted for education.’

In (19), the verb specifies the nature of the process by which more money is acquired. The semantic roles of the arguments remain the same regardless of the verb. Note, however, that this is not to say that (18) is an elliptic version of (19) but to illustrate differences between verbless and finite conceptualizations. In fact, as mentioned above, there are many possible ways to construe (18). On the other hand, certain instances of verbless constructions, especially those containing a deverbal NPgram and an OblP with a stative meaning, do name the type of the event — but they do so at the participants’ expense. Consider:



Semantic roles and verbless constructions

Finnish (Language Bank) (20) a. Sri Lanka-ssa kiiva-i-ta taistelu-j-a vaali-en alla Sri Lanka-ine fierce-pl-ptv battle-pl-ptv election-gen.pl under lit. ‘In Sri Lanka fierce battles before elections.’ b. Kuolonkolari Harjavalla-ssa fatal.car.accident Harjavalta-ine [town] lit. ‘A fatal car accident in Harjavalta.’

In (20a) (see also (14b)), the deverbal NPgram kiivaita taisteluja ‘fierce battles’ specifies the nature of the event, while the participants and the relations between them remain unspecified. The initial OblP Sri Lankassa ‘in Sri Lanka’ of course suggests that at least some of the participants involved are Sri Lankan, but the expression is primarily about naming the event, not about its participants. The internal structure of the battles is not described by the construction. According to the principles of construal stated in the previous section, the expression in (20a) approximates the meaning ‘Fierce battles are taking place before elections in Sri Lanka’ or ‘Fierce battles are fought before elections in Sri Lanka’. Furthermore, as shown by (20b), it is possible to form a similar conceptualization with an NPgram that is not deverbal but otherwise lexically specifies the sort of the event (see (14a)). Verbless constructions, then, represent events schematically in many respects: they specify neither the particular nature of the event nor its participants and the relations between them. The remaining information, however, is exactly the kind of information that is relevant with regard to semantic roles. Namely, the form of a construction indicates the perspective from which an event has been conceptualized, e.g. which participant is seen as the energy source, and both lexical and pragmatic factors contribute to disambiguating between semantic roles that share some features such as animacy. This is the topic of the next section. 4.3 Semantic roles attested in verbless constructions In Sections 4.1 and 4.2, we discussed how verbless constructions express events and what kind of information on events can be conveyed without verbs. In this section, we proceed to the main question of our study, i.e. what semantic roles are based on in verbless constructions and, possibly, in general as well. It is important to note that defining individual roles and their correct or sufficient number lies outside the scope of this chapter, although they are questions that deserve in-depth studies of their own. Instead, our discussion has potential consequences for theories of semantic roles, as we argue with the help of verbless data that no direct link between verbs and semantic roles is needed but that semantic roles are based on

97

98 Katja Västi and Seppo Kittilä

constructional semantics. The idea is closely related to Goldberg’s (1995, 2006) theory of argument structure constructions that differentiates between the meaning of a verb and that of a construction. Semantic roles (argument roles) are part of a construction’s meaning, whereas verbal semantics includes frame-specific participant roles (we use the concept of frame in the sense of Goldberg (e.g. 1995: 25–27, 2006: 38–40)). For instance, the meaning of the English ditransitive construction is analyzed as containing the argument roles of agent, patient, and (willing) recipient, and verbs with certain semantics — i.e. with certain kinds of participant roles — can be used in the construction. One of these verbs is hand, whose participant roles hander, handee, and handed are fused with the abovementioned semantic roles when used in the ditransitive construction (Goldberg 1995: 43–52, 141–151). It is not arbitrary as to which verbs may occur in which constructions, but there are principles guiding the fusion of participant roles and argument roles. Typically, constructions are polysemous, and their different senses are associated with different verb classes. The same largely applies to verbless constructions, except that there are usually several participant roles available for their arguments depending on the construal and, consequently, instances of verbless constructions tend to be ambiguous between different senses. That is, in a verbless construction, lexical elements can be construed as belonging to different semantic frames. Building on both our theoretical approach and data, we see constructional semantics as fundamental for associating particular semantic roles with particular verbless expressions; constructions have independent, relatively schematic meanings that determine semantic roles for their instances. This can be seen particularly clearly in the instances where a grammatical element bears an exceptional meaning in a verbless construction, and only in that construction. An example of this is the Finnish allative case which may indicate an agentive participant in a particular verbless construction (see Västi 2011a for a detailed discussion of the construction in question), but a similar meaning is not possible if a verb is present. Moreover, it is a highly atypical sense for any goal-marking morpheme (see Rice & Kabata 2007 for an extensive survey on their crosslinguistic grammaticalization patterns). Consider: (21)

Finnish (Language Bank) Lumpee-lle 100. NHL-maali Lumme-all hundredth NHL.goal [surname] ‘Lumme scored [his] hundredth NHL goal.’ (lit. ‘For Lumme the hundredth goal in NHL.’)

Based on our intuition, the only reasonable construal for (21) is that Lumme scores the goal as a result of his own intentional activity; it is not possible that the goal is scored by an unspecified player on behalf of Lumme, or that Lumme is a goalkeeper



Semantic roles and verbless constructions 99

who fails to make a save in the NHL for the hundredth time. However, the meaning of the expression involves the information that scoring a goal remains as an abstract achievement within Lumme’s sphere of control. We may thus suggest that the semantic role of the argument Lumpeelle is, say, agentive recipient. It is a highly idiosyncratic role, but defining Lumpeelle as a recipient, for instance, would leave out the clearly relevant semantic feature of agentivity. We may add that if (21) were rephrased with a finite construction, Lumpeelle would appear as a subject in the nominative case (i.e. Lumme) underlining its agentive nature (see Västi 2011a for paraphrase tests on verbless constructions with an initial-position allative element). If semantic roles are defined as feature bundles that are constructional attributes, then it follows that they are connected with certain morphosyntactic features. As repeatedly shown by cognitive-functional studies of grammar, the marking of lexical elements is semantically motivated, whether this is synchronically transparent or not. In other words, grammatical elements have meanings of their own, and especially senses of semantic cases and adpositions are relatively specific, at least when compared to grammatical cases. Accordingly, the recognition of the semantic roles of a construction is largely based on the form of its elements. The relationship between the form and the function of a construction is typically rather direct but, as shown by (21), constructions may be more idiomatic as well. A construction with more predictable semantics is illustrated in (22): Finnish (Language Bank) (22) Kapea systeemi Lahte-en narrow system Lahti-ill [town] lit. ‘A narrow system to Lahti.’

In (22), the semantic role of the NPgram depends partly on whether it is construed as a subject-like or an object-like argument, whereas there is only one available semantic role for the illative-marked argument, which could be called, for instance, goal. The meaning of the explicitly marked element is rather directly dictated by its morphological marking in cases like (22). Morphology alone does not, of course, provide sufficient information to construe semantic roles in verbless expressions but, just as with any construction, lexical semantics plays an important role. There are several lexical semantic features affecting semantic role assignment, such as animacy and abstractness vs. concreteness of the referent. Effects of animacy are exemplified in (23): Finnish (Language Bank) (23) a. Patosilla-lle takorautakaitee-t dam.bridge-all wrought.iron.railing-nom.pl lit. ‘To the dam bridge wrought iron railings.’

100 Katja Västi and Seppo Kittilä

b. c.

Minke-i-lle ehkä uima-allas mink-pl-all maybe swimming.pool lit. ‘To/for minks maybe a swimming pool.’ Maatalouskoulu-lle muistolaatta agricultural.school-all memorial.plaque lit. ‘To/for the agricultural school a memorial plaque.’

Examples in (23a–b) are formally similar but differ in the animacy of the allativemarked participant. Accordingly, the semantic role of the argument patosillalle ‘to a dam bridge’ in (23a) can be defined as goal and that of the argument minkeille ‘to/for minks’ in (23b) as recipient or beneficiary, depending on the construal. The example in (23c) is also formally comparable, but its allative-marked argument maatalouskoululle ‘to/for an agricultural school’ is ambiguous with regard to the referent’s animacy; it is possible to construe it as a concrete building or, metonymically, as a group of people constituting the institution. Consequently, maatalouskoululle may function as a recipient or a goal. Morphology cannot be responsible for the differences in (23), but only the animacy of the allative-marked participant provides us with a satisfactory result. In addition to constructional and lexical semantics, pragmatic and contextual factors also contribute toward construing semantic roles for verbless expressions. There is no verb specifying the event but several alternative semantic frames evoked by arguments together may be available, and the frames, for their part, may be compatible with different event types. In those cases, world knowledge and cultural knowledge influence choices among event types and, consequently, among semantic roles. That is, pragmatics influences the construal of a relevant event type for a verbless expression, and, as defined in Section 4.1, event types contain particular semantic roles. This is particularly relevant for those semantic roles that share one or more features (e.g. animacy, targetness). The situation is very different from constructions with verbs where the verbal semantics confines the possible variation. As mentioned above, Finnish allative-marked arguments may indicate agentive recipients when occurring in the initial position in a verbless construction (cf. (21)). Often, it is possible to construe the same arguments as mere recipients, which is the predictable scenario on the basis of the allative marking. However, pragmatic factors, such as cultural knowledge concerning referents, seem to increase or decrease the possibility that those arguments might function as agentive recipients. This is illustrated in (24a–b): Finnish (Language Bank, p.k.) (24) a. Mika Häkkise-lle mökkitontti Enontekiö-ltä Mika Häkkinen-all cabin.lot Enontekiö-abl [personal name] [town] lit. ‘For Mika Häkkinen a cabin lot from Enontekiö.’

Semantic roles and verbless constructions 101



b. Työttöm-i-lle mökkitontti Enontekiö-ltä unemployed-pl-all cabin.lot Enontekiö-abl [town] lit. ‘For the unemployed a cabin lot from Enontekiö.’

The corpus example (24a) can refer both to an event where Mika Häkkinen buys the cabin lot for himself or also to an event where he receives the lot as a gift. That is, the argument Mika Häkkiselle ‘to Mika Häkkinen’ can function as a recipient or as an agentive recipient. Both construals are pragmatically plausible: as Mika Häkkinen is a famous (former) Formula 1 driver and thus wealthy, he is an excellent candidate to purchase a cabin lot, and as in Finland lots are donated to successful sportsmen, he is at least equally likely to receive one given to him. The modified elicited example (24b), for its part, does not easily allow for the agentive recipient reading, even though it is identical in form with (24a). One probable reason for this is that it is culturally unlikely that unemployed people purchase cabin lots. However, in principle nothing in the form of the expression prevents the agentive recipient construal. In the context of examples (23a–c), we discussed the role of lexical semantics for the construal of semantic roles focusing on a single element at a time. However, it is important to recognize the role of lexical context as well. In verbless constructions, the semantic role of an argument often depends strongly on the meaning of the accompanying argument: cases and adpositions tend to be polysemous, and individual lexical elements can appear in countless semantic frames, so the function of an argument may be contextually determined. Consider (25) where the initial-position argument occurs in the elative case: (25)

Finnish (Language Bank) Katajanpuisto-sta leikkipuisto Kataja.park-ela playground [juniper’s park] lit. ‘From Kataja Park a playground.’

The elative is a highly polysemous case in Finnish (see Leino 1993), but the concrete spatial sense ‘from [somewhere]’ can be seen as the central one. This sense would be compatible with the meaning of the stem Katajanpuisto- ‘Kataja Park’ but not with the meaning of the accompanying argument leikkipuisto ‘playground’; a playground cannot concretely come out of a park. Thus the argument Katajanpuistosta does not function as a physical source. The Finnish elative is used to indicate sources of more abstract changes as well, and (25) can be interpreted accordingly as, for instance, ‘Katajanpuisto became a playground’ or ‘Katajanpuisto was turned

102 Katja Västi and Seppo Kittilä

into a playground’. The best label for the semantic role of the elative argument is unclear to us, but the existence of the role can be justified nonetheless. Moreover, adjuncts in verbless expressions may also affect the semantic roles arguments receive. In fact, their function appears to be more crucial in verbless constructions than in full clauses containing a finite verb. Consider the examples in (26a–b): Finnish (Language Bank, p.k.) (26) a. Kiinteistövero vain väkisin ylös real.estate.tax only by.force up lit. ‘The real estate tax up only by force.’ b. Kiinteistövero ylös real.estate.tax up lit. ‘The real estate tax up.’

In principle, the corpus example (26a) corresponds to most instances of verbless constructions discussed above in that it is possible to construe the NPgram kiinteistövero ‘real estate tax’ at least as a subject-like or an object-like argument. That is, (26a) can be interpreted as, for instance, ‘The real estate tax goes up only by force’ or ‘The real estate tax is raised only by force’. However, the particle vain ‘only’ and the adverb väkisin ‘by force’ together suggest that the denoted event is caused by an agent, which emphasizes the latter construal. A modified example (26b), which does not include any elements besides the core arguments of the construction, in turn, does not lay emphasis on either construal. In constructions with a verb, adverbs are less important for our understanding of semantic roles, because the verb disambiguates between the possible readings, even though agentive adjuncts, such as ‘with intent’ and ‘carefully’, may in certain cases disambiguate between, for example, anticausative and passive readings. Finally, when dealing with verbless constructions, situational context also necessarily plays a role in assigning semantic roles. Consider examples (27a–b) (not included in our data but spotted in the newspaper Kaleva): Finnish (headlines, Kaleva) (27) a. Nokia House-lle rakennusluva-t Peltola-an Nokia House-all planning.permission-nom.pl Peltola-ill [district] lit. ‘For Nokia House planning permissions to Peltola.’ b. Kolme-lle sude-lle kaatolupa three-all wolf-all shooting.license lit. ‘For three wolves a shooting license.’



Semantic roles and verbless constructions 103

At first glance, examples in (27) seem to be instances of the same verbless construction composed of an initial-position allative-marked OblP and an NPgram. In addition, in both cases the NPgram is a compound word with the head lupa ‘permission, license’. Nonetheless, these instances do not allow for completely same sets of construals. One of them for (27a) is ‘Nokia House [as a metonymically conceptualized community] gets permission to build [something] in Peltola’, whereas, in a newspaper context, (27b) cannot be construed as ‘Three wolves get permission to shoot [something]’. Instead, it can be interpreted as, for example, ‘A shooting license for three wolves is granted [to unspecified persons]’, in which case (27b) is not an instance of the same construction as (27a). Yet, the expression does not in itself block the recipient construal of wolves but, in a fairytale context for instance, could denote an event in which wolves receive a shooting license. 5. Discussion In this chapter, we have discussed the nature of genuinely verbless constructions in light of Finnish data. We have shown that semantic roles can be seen as a part of the meaning of a construction and that they are directly motivated by the construction’s form, although lexical semantics plays an important role as well. Moreover, different pragmatic factors contribute to the construal of the roles for particular expressions. The relations between form and function are naturally language-specific, and not all languages may deal with verbless expressions equally well. We believe that semantic roles should be defined as relatively coarse-grained feature bundles tied to event types — semantic units that correspond to the kind of information that arguments of verbless constructions are able to provide — and that they exist independently of verbs (cf. Croft 1991; DeLancey 1991; Goldberg 1995, 2006; Næss 2003, 2007 whose proposals share features with ours). We may say that, in verbless constructions, it is the semantic layer of participant roles that is not explicated. They are nevertheless inferable: lexical elements of verbless expressions together evoke semantic frames within which participant roles are construed, along with semantic roles. Construals of verbless expressions, however, remain relatively fuzzy in any case, as nominal elements alone simply cannot convey all the information that a verb could. We have also discussed the fact that verbless constructions can be analyzed as polysemous, similarly to argument structure constructions in the spirit of Goldberg (1995, 2006). Consequently, commonly arising alternative sets of semantic roles in verbless constructions can be associated with their different senses. It is thus a matter of context as to which senses are represented by individual instances on individual occasions. That is, the systematic ambiguity of particular instances of

104 Katja Västi and Seppo Kittilä

verbless constructions is not an indication of undetermined semantics but of the regular polysemy of constructions. Certainly, the polysemy of verbless constructions is different from that in the argument structure constructions investigated by Goldberg in that verbless constructions allow variation in the perspective of construal — e.g. in construing the NPgram as subject-like or object-like — and not only in the type of the denoted process. The existence of alternative construals does not contradict the viewing of constructions as being responsible for semantic roles. As the brief discussion above implies, the most important finding of our paper is that verbs are not necessary for the definition of semantic roles, but roles similar to those attested in constructions with a verb are also found in genuinely verbless constructions. Dynamic meanings of verbless constructions are based on meaningful elements other than verbs, namely cases, adpositions, and adverbs, which function much like verbs as predicates. In the presence of such elements, the contribution of verbs to the reading of clauses is partially redundant, as the discussion of examples such as (19) has shown. In these cases, the function of the verb is rather to specify the nature of the process, and the semantic roles follow from the semantics of the constructions in question. This, of course, calls into question whether verbs should still be seen as the central elements of constructions, or whether the construction as a whole is indeed more important with regard to defining semantic roles. In our view, verbs contribute to the overall interpretation of the constructions but are not primarily responsible for the semantic role assignment. What implications this has for our understanding of the concept of semantic roles remains to be seen. We hope that our study is a first step towards numerous similar investigations of verbless constructions in other languages. Studying verbless constructions in formally and genealogically diverse languages would certainly make an important contribution toward defining the concept of semantic roles. In the following, we list some of the topics that we have not been able to address in this study but which are in need of closer research both in and across languages. First, it is important to study what features semantic roles necessarily consist of and to consider whether some of them should be seen as essential and some more peripheral. In other words, do the features form any kind of hierarchy, in which certain features are more relevant than others? Are the features recognized with the help of verbless constructions that should specifically be integral parts of a definition of semantic roles? As our data has shown, similar features, such as agency, are relevant to the definition of semantic roles regardless of whether a given construction includes a verb or not. This question is central to our understanding of semantic roles, as verbs have played a leading role in definitions thus far, while our study suggests that their importance has been exaggerated. Second, the crosslinguistic limits of verblessness are in need of closer scrutiny. Our study is based on one language and one type of data only, and we are looking

Semantic roles and verbless constructions 105



forward to seeing in which contexts verbless constructions occur and how other languages deal with the lack of verbs in similar constructions. Verbless constructions, just like constructions with a verb, are definitely different in nature in different languages, and by examining structurally diverse languages we could obtain a better picture of the roles and features that are generally possible, regardless of whether a given construction involves a verb or not. Finnish is a language with a rather rich case inventory, which may contribute in an important way to the lack of verbs, but studies of other languages are needed in this respect as well. Are verbs, for example, more important and thus less frequently absent in languages that lack case markers and adpositions, or do contextual clues gain more relevance in such cases? What are the limits of verblessness; what kinds of verbless constructions are possible across languages? Finally, one important issue that emerged from the discussion of cases such as (21) and (24) is that, in verbless constructions, grammatical elements, in this case the allative, have meanings they do not have in finite constructions. In finite constructions, the functions of the Finnish allative are typically related to motion, reception, and experience. However, in verbless constructions, a rather frequent semantic role of the allative is agentive recipient. One of the reasons for this atypical use of the allative in this particular verbless construction possibly lies in the agentive traits of the (willing) recipients also present in typical transfer events. In transfer events, the recipient needs to complete the event by accepting the transfer, which accords it certain agentive features. Furthermore, the agent is often unspecified in such cases. Are there comparable phenomena in other languages as well? Studying them could contribute to our understanding of semantic roles from a novel perspective, as unexpected traits may become relevant in verbless constructions and thus demonstrate the significance of constructional meaning.

Abbreviations abl acc ade adel all aor ela erg ess

Ablative Accusative Adessive Adelative Allative Aorist Elative Ergative Essive

gen gram ill ind ine inf InfP loc nom

Genitive Grammatical case Illative Indicative Inessive Infinitive Infinitive phrase Local case Nominative

NP OblP pass pl pp prs pst ptv sg

Noun phrase Oblique phrase Passive Plural Adpositional phrase Present tense Past tense Partitive Singular

106 Katja Västi and Seppo Kittilä

Acknowledgements We would like to thank the two anonymous referees and a member of the Studies in Language editorial board for their invaluable feedback on earlier versions of this chapter. We also thank the Academy of Finland (project number 1127724) for providing funding for this study. As usual, any remaining flaws are solely our responsibility.

References Åkermalm, Åke. 1965. Rubriksvenska och andra studier (Skrifter utgivna av Modersmålslärarnas Förening 100). Falköping: Gleerups. Conon, Lars. 1973. Rubrikspråket på Dagens Nyheters förstasidor I–II (Acta Universitatis Upsaliensis, Studia Philologiae Scandinavicae Upsaliensia 8). Uppsala: Uppsala Universitet. Croft, William. 1991. Syntactic categories and grammatical relations: The cognitive organization of information. Chicago & London: The University of Chicago Press. DeLancey, Scott. 1984. Notes on agentivity and causation. Studies in Language 8(2). 181– 213.  doi: 10.1075/sl.8.2.05del DeLancey, Scott. 1991. Event construal and case role assignment. Seventeenth Annual Meeting of the Berkeley Linguistics Society (BLS). 338–353. Dixon, R. M. W. 2010. Basic Linguistic Theory: Volume 1 methodology. Oxford: Oxford University Press. Dowty, David. 1989. On the semantic content of the notion ‘thematic role’. In Gennaro Chierchia, Barbara H. Partee & Raymond Turner (eds.), Properties, types and meaning II: Semantic issues (Studies in Linguistics and Philosophy), 69–129. Dordrecht: Kluwer.  doi: 10.1007/978-94-009-2723-0_3

Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67(3). 547– 619.  doi: 10.1353/lan.1991.0021 Eriksen, Pål, Seppo Kittilä & Leena Kolehmainen. 2010. The linguistics of weather: Crosslinguistic patterns of meteorological expressions. Studies in Language 34(3). 565–601.  doi: 10.1075/sl.34.3.03eri

Fillmore, Charles J. 1968. The case for case. In Emmon Bach & Robert T. Harms (eds.), Universals in linguistic theory, 1–88. London: Holt, Rinehart and Winston. Fillmore, Charles J. 1970. The grammar of hitting and breaking. In Roderick A. Jacobs & Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar, 120–133. Waltham: Ginn. Fried, Mirjam & Jan-Ola Östman. 2004. Construction Grammar: A thumbnail sketch. In Mirjam Fried & Jan-Ola Östman (eds.), Construction Grammar in a cross-language perspective (Constructional Approaches to Language 2), 11–86. Amsterdam & Philadelphia: John Benjamins.  doi: 10.1075/cal.2.02fri Goldberg, Adele E. 1995. Constructions: A Construction Grammar approach to argument structure (Cognitive Theory of Language and Culture). Chicago & London: The University of Chicago Press. Goldberg, Adele E. 2006. Constructions at work: The nature of generalization in language. Oxford: Oxford University Press.



Semantic roles and verbless constructions 107

Gruber, Jeffrey. 1965. Studies in lexical relations. Cambridge, MA: MIT dissertation. Haspelmath, Martin. 1993. A grammar of Lezgian (Mouton Grammar Library 9). Berlin & New York: Mouton de Gruyter.  doi: 10.1515/9783110884210 Helasvuo, Marja-Liisa. 1991. Syntaktiset rakenteet kertomuksen jäsennyksen osoittimina. Virittäjä 95(1). 33–47. Helasvuo, Marja-Liisa. 2001. Syntax in the making: The emergence of syntactic units in Finnish conversation (Studies in Discourse and Grammar 9). Amsterdam & Philadelphia: John Benjamins.  doi: 10.1075/sidag.9 Huumo, Tuomas. 2003. Incremental existence: The world according to the Finnish existential sentence. Linguistics 41(3). 461–493.  doi: 10.1515/ling.2003.016 Huumo, Tuomas & Krista Ojutkangas. 2006. An introduction to Finnish spatial relations: Local cases and adpositions. In Marja-Liisa Helasvuo & Lyle Campbell (eds.), Grammar from the human perspective: Case, space and person in Finnish (Current Issues in Linguistic Theory 277), 11–20. Amsterdam & Philadelphia: John Benjamins.  doi: 10.1075/cilt.277.04huu Jackendoff, Ray. 1972. Semantic interpretation in Generative Grammar (Current Studies in Linguistics 2). Cambridge, MA: MIT Press. Jackendoff, Ray. 1976. Toward an explanatory semantic representation. Linguistic Inquiry 7. 89–150. Kittilä, Seppo. 2005. Remarks on involuntary agent constructions. Word 56(3). 381–419. Kittilä, Seppo. 2008. Animacy effects on Differential Goal Marking. Linguistic Typology 12(2). 245–268.  doi: 10.1515/LITY.2008.038 Kittilä, Seppo & Jussi Ylikoski. 2011. Remarks on the coding of Direction, Recipient and Vicinal Direction in European Uralic. In Seppo Kittilä, Katja Västi & Jussi Ylikoski (eds.), Case, animacy and semantic roles (Typological Studies in Language 99), 29–64. Amsterdam & Philadelphia: John Benjamins. Kopotev, Mikhail. 2007a. What a difference a verb makes! Russian and Finnish verbless sentences. In Marja Nenonen & Sinikka Niemi (eds.), Collocations and idioms 1: Papers from the First Nordic Conference on Syntactic Freezes, Joensuu, May 19-20, 2006 (Studies in Languages, University of Joensuu, vol. 41), 177–192. Joensuu: Joensuu University Press. Kopotev, Mikhail. 2007b. Where Russian syntactic zeros start: Approaching Finnish? In Juhani Nuorluoto (ed.), Topics on the ethnic, linguistic and cultural making of the Russian North (Slavica Helsingiensia 32), 116–137. Helsinki: University of Helsinki. Lazard, Gilbert. 1998. Actancy. Berlin & New York: Mouton de Gruyter.  doi: 10.1515/9783110808100 Leino, Pentti. 1993. Polysemia – kielen moniselitteisyys: Suomen kielen kognitiivista kielioppia 1 (Kieli 7). Helsingin yliopiston suomen kielen laitos. Leino, Pentti. 2001. Verbit, konstruktiot ja lausetyypit. In Pentti Leino, Ilona Herlin, Suvi Honkanen, Lari Kotilainen, Jaakko Leino & Maija Vilkkumaa, Roolit ja rakenteet: Henkilöviitteinen allatiivi Biblian verbikonstruktioissa (Suomalaisen Kirjallisuuden Seuran Toimituksia 813), 11–66. Helsinki: Suomalaisen Kirjallisuuden Seura. Moor, Marianne. 1985. Studien zum lesgischen Verb. Wiesbaden: Otto Harrassowitz. Mårdh, Ingrid. 1980. Headlinese: On the grammar of English frontpage headlines (Lund Studies in English 58). Lund: CWK Gleerup. Næss, Åshild. 2003. Transitivity: From semantics to structure. Nijmegen: University of Nijmegen dissertation. Næss, Åshild. 2007. Prototypical transitivity (Typological Studies in Language 72). Amsterdam & Philadelphia: John Benjamins.  doi: 10.1075/tsl.72

108 Katja Västi and Seppo Kittilä Newmeyer, Frederick J. 2010. On comparative concepts and descriptive categories: A reply to Haspelmath. Language 86(3). 688–695.  doi: 10.1353/lan.2010.0000 Payne, Thomas E. 1997. Describing morphosyntax. Cambridge: Cambridge University Press. Rappaport Hovav, Malka & Beth Levin. 2008. The English dative alternation: The case for verb sensitivity. Journal of Linguistics 44. 129–167. Reinhart, Tanya. 1996. Syntactic effects of lexical operations: Reflexives and unaccusatives. OTS Working Papers in Linguistics. University of Utrecht: Utrecht Institute of Linguistics. Reinhart, Tanya. 2000. The theta system: Syntactic realization of verbal concepts. OTS Working Papers in Linguistics. University of Utrecht: Utrecht Institute of Linguistics. Reinhart, Tanya. 2001. Experiencing derivations. In Rachel Hastings, Brendan Jackson & Zsofia Zvolenszky (eds.), Semantic and Linguistic Theory (SALT 11), 365–387. Ithaca: Cornell Linguistics Circle Publications. Reinhart, Tanya. 2002. The theta system – an overview. Theoretical Linguistics 28(3). 229–290. Rice, Sally & Kaori Kabata. 2007. Crosslinguistic grammaticalization patterns of the allative. Linguistic Typology 11(3). 451–514.  doi: 10.1515/LINGTY.2007.031 Rozwadowska, Bożena. 1988. Thematic restrictions on derived nominals. In Wendy Wilkins (ed.), Syntax and semantics 21: Thematic relations, 147–165. New York: Academic Press. Rozwadowska, Bożena. 1989. Are thematic relations discrete? In Roberta Corrigan, Fred Eckman & Michael Noonan (eds.), Linguistic categorization (Current Issues in Linguistic Theory 61), 115–130. Amsterdam & Philadelphia: John Benjamins.  doi: 10.1075/cilt.61.09roz Schneider, Kristina. 2000. The emergence and development of headlines in British newspapers. In Friedrich Ungerer (ed.), English media texts – past and present: Language and textual structure (Pragmatics & Beyond New Series 80), 45–65. Amsterdam & Philadelphia: John Benjamins.  doi: 10.1075/pbns.80.05sch Simon-Vandenbergen, Anne-Marie. 1981. The grammar of headlines in The Times 1870–1970 (Verhandelingen van de Koninklijke Academie voor Wetenschappen, Letteren en Schone Kunsten van België 95). Brussels: Paleis der Academiën. Siro, Paavo. 1964. Suomen kielen lauseoppi. Helsinki: Tietosanakirja. Song, Jae Jung. 2011. There’s more than “more animate”: The Organization/Document Construction in Korean. In Seppo Kittilä, Katja Västi & Jussi Ylikoski (eds.), Case, animacy and semantic roles (Typological Studies in Language 99), 183–206. Amsterdam & Philadelphia: John Benjamins.  doi: 10.1075/tsl.99.07son Straumann, Heinrich. 1935. Newspaper headlines: A study in linguistic method. London: George Allen & Unwin. Vahtera, Ralf. 2009. Otsikkorakenteita: Kontrastiivinen tutkimus suomen- ja ruotsinkielisten sanomalehtien syntaktis-semanttisista otsikkotyypeistä (Acta Wasaensia No 209, Kielitiede 41). Vaasa: University of Vaasa. Van Valin, Robert D., Jr. & Randy J. LaPolla. 1997. Syntax: Structure, meaning and function (Cambridge Textbooks in Linguistics). Cambridge: Cambridge University Press.  doi: 10.1017/CBO9781139166799

Västi, Katja. 2011a. A case in search of an independent life: Semantics of the initial allative in a Finnish verbless construction. In Seppo Kittilä, Katja Västi & Jussi Ylikoski (eds.), Case, animacy and semantic roles (Typological Studies in Language 99), 65–109. Amsterdam & Philadelphia: John Benjamins.  doi: 10.1075/tsl.99.03vas Västi, Katja. 2011b. Mihin verbittömien konstruktioiden merkitystyypit perustuvat? Skemaattiset ja polyseemiset tapahtumanilmaukset. Sananjalka 53. 34–60. Västi, Katja. 2012. Elävä lähde: Alkuasemaisen ablatiivin merkitystyypit verbittömässä konstruktiossa. Virittäjä 116(1). 67–97.

Benefaction proper and surrogation Fernando Zúñiga University of Bern

The semantic role of beneficiary is usually conceptualized in very general terms, typically without an intensional definition of what can constitute a benefit in the particular construction under study. Among those accounts that have proposed to discuss benefaction as related to the notion(s) of surrogation, substituting, and/or deputing, Kittilä (2005) proposes a distinction between recipients, beneficiaries, and recipient-beneficiaries based on the binary features [reception] and [substitutive benefaction]; the recipient includes only reception (and the beneficiary only substitutive benefaction), whereas both features are relevant with recipient-beneficiaries. This chapter proposes an alternative account (i) by defining benefaction proper in terms of a prototype related to possession (and thereby to reception) and a periphery, and (ii) by defining surrogation as a separate notion that can, but need not, coalesce with benefaction proper. Thus, the beneficiaries’ condition improves because they are relieved from having to carry out a given action themselves. Keywords: beneficiary, benefactives, surrogation, deputative beneficiary, substitutive beneficiary

1. Introduction Most accounts of the thematic relation usually called beneficiary conceptualize this notion in rather general terms, viz. as a “generally […] animate [being] indirectly affected by the action” (Palmer 1994: 10), often providing only a handful of English examples like the boy bought the girl a book, and without a detailed discussion of what kinds of benefits can be expressed by the particular constructions that serve as illustrations. Occasionally, accounts along quite different lines are found as well: “the animate entity on whose behalf an activity is carried out, [as in] she did the shopping for her mother” (Blake 1994: 70). Some studies have proposed to discuss benefaction (typically, a general equivalent of English ‘for’) as explicitly associated with the notions of surrogation (‘on behalf of ’) and substitution (‘instead of ’), and it is the connection between benefaction and these related notions that doi 10.1075/bct.88.05zun 2016 © John Benjamins Publishing Company

110 Fernando Zúñiga

is at the center of attention in the present article. I would like to claim that what I call benefaction proper here is best seen as different from both surrogation and substitution, and that some languages provide evidence suggesting that an adequate account of these notions must consider them as such despite the fact that they can be expressed through the same formal means (either in those very same languages or in others). This chapter elaborates on ideas already formulated in Zúñiga (2011a, 2011b), without the formalism introduced there but with a special emphasis on the empirical evidence. A note on terminology and analysis is in order before proceeding. I utilize the labels thematic relation, semantic role, and participant role interchangeably in what follows. I use the term beneficiary to denote a participant role and the label benefactive to refer to a formal means employed to express that a given participant benefits from a state of affairs. (In Section 2, when reviewing those authors’ views on the topic who use benefactive to denote the thematic relation, I make it clear that it is their usage by using double inverted commas when appropriate.) I do not want to argue in favor of a particular account of all thematic relations but merely to argue that, whatever theory of semantic roles an analyst ends up endorsing, it will be most useful if it takes into account the notion of surrogation along lines similar to the ones proposed here. Section 2 reviews some previous studies of surrogation, while Section 3 presents the view of the different subtypes of benefaction espoused here. Section 4 surveys three case studies of benefactive constructions that support such a view, and Section 5 closes the chapter with some conclusions and suggestions for further research. 2. Previous studies of surrogation In his work on participant roles, Frawley characterizes his “benefactives” as participants that “derive actions or entities from the actions of another” (1992: 215) and mentions two possible readings of the English utterance Dr. Frankenstein made his son a monster. On one reading (‘Dr. Frankenstein converted his son into a monster’), his son is a patient (a different participant role in Frawley’s model), while on the other (‘Dr. Frankenstein made a monster for his son’), his son “comes into the possession of something because the actions of Dr. Frankenstein bring it about” and has “a benefactive relation to the predicate” (1992: 216). Interestingly enough, Frawley says that “[b]enefactives may be viewed, alternatively, as arguments inherently dependent on surrogates” (p. 216). In the benefactive reading of Dr. Frankenstein made his son a monster, “Dr. Frankenstein is a surrogate, or substitute performer.” In some cases, “[t]he surrogacy may even be so

Benefaction proper and surrogation 111



strong as to preempt or preclude action by the benefactive,” e.g. Tom lost the game for his team and Mary bought lunch for Bob (p. 216). The action carried out by Tom affects all members of the team in the former and Mary carries out the act and prevents Bob from paying in the latter. Frawley further says that such co-optation is an example of extreme surrogacy but “not a necessary feature of benefactives” (p. 216). To be sure, such a view is not the first account of such different readings of English constructions based on for. According to Somers’ survey of several influential studies from the 1970’s that addressed the beneficiary as a semantic role, the ‘Benefactive’ case is partially introduced by Fillmore [(1968: 32)], though without any clear indication of its characterisation, other than its association with the preposition for. The case as used by other authors seems to have two distinct meanings, one having to do with possession, and the other in instances where the paraphrase on behalf of is appropriate. (Somers 1987: 133)

He concludes that the former meaning can be summarized as covering the gain (or loss) of possession, as well as nondynamic possession. Somers then presents two approaches to the ‘on behalf of ’ meaning. One can be seen in the treatment by authors like Teng (1975) in terms of “two distinct types,” viz. intended reception and substitution, which receive different formal expression in Mandarin (1a and 1b respectively): (1) Mandarin (Teng 1975: 151) a. Wo3 gei3 ta1 xie3 xin4. 1 give 3 write letter b. Wo3 ti4 ta1 xie3 xin4. 1 replace 3 write letter Both: ‘I’ll write a letter for him/her.’

The other approach is Platt’s (1971: 50) analysis in terms of “inner Benefactives,” which are recipients and possessors, and “outer Benefactives,” which cover “persons who benefit generally from the action of the verb” without there being any “necessary notion of eventual possession or receipt” (Somers 1987: 135). Lastly, Somers says that “[i]t is not entirely clear whether these three notions of Benefactive [i.e. Fillmore’s, Teng’s, and Platt’s, FZ] are mutually exclusive” (1987: 135). He regards an English sentence like (2) in the context of Mr. Jones (the substitute) asked me to write a letter to the newspaper (the recipient) for Mr. Smith (the general 1.  This formulation owes much to ideas expounded in a programmatic article by Shibatani (1996) and a case study by Song (2010) as to the exact role played by possession in some East Asian ‘give’-benefactives. See Subsection 3.1 (towards the end) for more details on Shibatani’s view, and Zúñiga (2011a) for a detailed discussion of the analytical issues involved.

112 Fernando Zúñiga

beneficiary) as admittedly marginal but not unacceptable and hypothesizes that “outer Benefactives” (like outer roles in general) can probably stack up “as far as processing limitations allow” (p. 136). (2) English (Somers 1987: 136) I’m writing the newspaper a letter for Mr. Smith for Mr. Jones.

Within the same tradition, Van Valin & LaPolla (1997) distinguish between recipient-beneficiaries (3a′), deputative beneficiaries (3a′′), and plain beneficiaries (3b): (3) English (Van Valin & LaPolla 1997: 383f) a. Robin baked a cake for Sandy. Reading a′: Roughly equivalent to Robin baked Sandy a cake. Reading a′′: Roughly equivalent to Robin baked a cake instead of Sandy. b. Rita sang for the students. Default reading: ‘Rita sang for the students’ enjoyment / to entertain the students.’

Terminologically, these are equivalent to (a subtype of) Teng’s recipients and Platt’s inner Benefactives (3a′) on the one hand and to the two kinds of outer Benefactives on the other (substitutes in (3a′′) and general beneficiaries in (3b)). Analytically, however, such participant roles are defined in a particular way within the Role and Reference Grammar framework, viz. resorting to a specific way of modeling the semantic structure of arguments, their arguments, and adjuncts based on a semantic decomposition model. Elaborating on Van Valin & LaPolla’s classification, Kittilä (2005) proposes a distinction between recipients, beneficiaries, and recipient-beneficiaries based on the notions of reception and substitutive benefaction (and as opposed to a different beneficiary subtype based on concrete benefaction). These notions are summarized in Table 1 and exemplified in Examples (4) through (5). Table 1.  Selected semantic roles according to Kittilä (2005) reception

substitutive benefaction

concrete benefaction (5)

a. Recipient (4a)

+

–

n/r

b. Beneficiary (4b)

–

+

n/r

c.

+

+

n/r

Recipient-beneficiary (4c)

(4)

English (Kittilä 2005: 273f) a. The teacher gave/sent me a book. b. The teacher parked the car for me / on my behalf. c. She baked a cake for me / baked me a cake.

Benefaction proper and surrogation 113



(5) English “concrete benefaction” (Kittilä 2005: 273f) a. The professor built a house for me (e.g. to occasionally accommodate me). b. The teacher lied for me (e.g. so that I will avoid punishment).

Armed with these analytic tools, Kittilä then classifies languages based on the construction inventory they display and the neutralizations they make. If a language consequently distinguishes the formal expression of all three notions of recipient, beneficiary, and recipient-beneficiary, it illustrates the tripartite type (6a). If, on the contrary, two of these notions are marked alike, it is labeled recipient-prominent if recipients and recipient-beneficiaries are not distinguished (6b) or beneficiaryprominent if the converse is the case (6c). Lastly, the neutral type (6d) is instantiated by those languages where all three roles are marked alike: (6)

Typology of benefaction marking according to Kittilä (2005) a. Tripartite (rec – ben – rec-ben) b. Recipient-prominent (rec/rec-ben – ben) c. Beneficiary-prominent (ben/rec-ben – rec) d. Neutral (rec/ben/rec-ben)

The different views on surrogation outlined to date can be summarized as in Table 2 below. Frawley (1992) seems to see surrogacy, at least in some instances of the English double-object construction (DOC) and for-constructions, as an alternative characterization of the same beneficiary role. Teng (1975) sees the notion related to intended reception and the one related to substitution as representing distinct subtypes of the beneficiary in the Mandarin serial constructions. Platt’s (1971) inner beneficiaries are roughly equivalent to Teng’s first subtype, but his outer beneficiaries cover those instances where the benefit obtained does not derive from substitution or surrogacy. Van Valin & LaPolla (1997) postulate three distinct subtypes of beneficiaries, with their plain beneficiary type being roughly equivalent to the non-deputative subset of Platt’s outer beneficiaries. Lastly, Kittilä (2005) actually proposes three distinct underlying notions (viz. (i) benefit as derived from surrogacy, (ii) benefit only generally or indirectly linked to the event/ Table 2.  Previous views of benefaction and surrogation intended reception/ possession

surrogacy

general benefit

Frawley (1992)

bena

benb

–

Teng (1975)

ben1

ben2

–

Platt (1971)

inner ben

outer ben

Van Valin & LaPolla (1997)

recipient-ben

deputative-ben

plain ben

Kittilä (2005)

recipient-ben

substitutive ben

concrete ben

114 Fernando Zúñiga

action, and (iii) reception), and his beneficiary subtypes are based on them: only (i) for his substitutive beneficiary, both (i) and (iii) for his recipient-beneficiary, and only (ii) for his concrete beneficiary. Despite the usefulness and formal appeal of Kittilä’s typology, the view of benefaction presented in the present paper differs slightly from it. Rather, it is an elaboration of the following skeptical remark by Jae Jung Song: “It seems that substitutive benefaction is not a basic conceptual notion to be used for the understanding of benefaction, but it is a proper type of benefaction in its own right” (2010: 397). The following section argues that such a view is indeed more adequate and more useful for the study of benefaction. 3. Benefaction proper vs. surrogation The accounts of benefaction outlined in the preceding section raise several questions, but I will limit myself to addressing the following three issues here. First, what is the exact relation between substitutive benefaction (she parked the car for me / on my behalf) and mere substitution (she parked the car instead of me)? Second, in Van Valin & LaPolla’s semantic decomposition of the different benefaction subtypes, deputative benefaction is one alternative among others. For Kittilä (2005), however, deputative benefaction underlies all nonconcrete benefaction subtypes. Third, the two examples of Kittilä’s concrete benefaction in (5) are quite different from each other: according to Kittilä, in (5a) the beneficiary is neither the intended nor the eventual recipient/possessor but benefits from the event nonetheless. In (5b), there is no necessary relationship whatsoever to reception/possession. The analysis advanced here will help address these three issues in a principled way. I propose to articulate a view of benefaction as complex of related meanings grouped in two subdomains, viz. benefaction proper (§3.1) and surrogation (§3.2). 3.1 Benefaction proper as prototype The prototype of benefaction proper can be defined as a bieventual complex event: An agent acts on a patient, thereby advantageously affecting the beneficiary, who is in a position to (better) interact with it as a consequence. This advantageous position may be due to the beneficiary obtaining permanent possession over the patient, but actually temporary possession (crucially defined by control, as in the intruder had a gun) seems to not only be enough but also the default. (In other words, I have not found in the languages of the world benefactive constructions whose applicability is restricted by ownership, whatever its definition might



Benefaction proper and surrogation 115

be in the relevant cultural setting.)1 Syntactically, the only requisite is that the benefactive be a monoclausal construction.2 Based on this definition, two subtypes can be distinguished at the core of the notion. In the first subtype, the beneficiary is the possessor of the patient before the state of affairs depicted by the predicate holds, and the beneficial act performed by the agent does not involve reception, i.e., the patient is simply an affected possessum (7a). In the second subtype, the beneficiary is the intended, albeit not necessarily the actual, recipient of a patient that has come into being as a result of the act performed by the agent, i.e., the patient is effected rather than merely affected (7b): (7) English (p.k.) a. She cleaned my basket for me. b. She wove me a basket.

There are several parameters along which a state of affairs can deviate from the prototype just sketched. First, with respect to a dimension that can be labeled involvement, the benefit may be derived not from the possibility to interact with the patient in some unspecified but pragmatically licensed way as in (7) (which I will call object benefaction), but rather from the event as a whole — possibly through a more specific kind of interaction on the part of the beneficiary, viz. via a psych predicate (8). I will call this subtype event benefaction:3 (8) English (p.k.) a. She sang for me (to hear). b. The teacher built a house for me (to watch how he did it).

By a similar token, what can be termed absolute benefaction consists in the beneficiary deriving a benefit from the state of affairs as a whole as well, but indirectly rather than through direct involvement with it (9): (9) English (p.k.) The teacher lied for me. (= 5b above)

Based on the prototype definition above and the distinctions along the parameter of involvement, the last question formulated at the outset of this subsection can 2.  Instead of “complex event,” I used the term “macro-event” in a Talmy-like broad sense in Zúñiga (2011a, 2011b). It is important to note that the notion at stake here crucially differs from what Bohnemeyer et al. (2007) and most articles in Bohnemeyer & Pederson (2011) call macro-event. With the latter, time-positional and durational operators (tense markers, temporal adverbs/clauses, etc.) have scope over all the entailed subevents; complex events as conceptualized here do not have this property. 3.  Here I owe, both terminologically and analytically, a substantial debt to Smith (2005). See also Hole (2008: 131) for the distinction of object benefaction and event benefaction.

116 Fernando Zúñiga

be addressed in a straightforward manner. The difference between the two distinct cases of Kittilä’s (2005) concrete benefaction in the teacher built a house for me (e.g. to occasionally accommodate me) and the teacher lied for me (e.g. so that I will avoid punishment) lies in the fact that the former corresponds to a specific reading of an utterance expressing object benefaction (one that contextually rules out exclusive permanent possession) while the latter is a case of absolute benefaction. There are several additional ways in which benefaction proper can deviate from the prototype, and I will limit myself to mentioning them briefly here for the sake of completeness. One of these dimensions bears relation to the kinds of predicates involved in the causing and the resulting subevents. In a prototypical benefaction situation, the actions of an agent will be telic, dynamic, etc., and the possible interactions on the part of a beneficiary might be socially sanctioned by custom. In less prototypical benefaction proper, the causing participant may be less than agentive and responsible for atelic, stative states of affairs; in some languages of North and South America, for example, applicative benefactives can be based on looking or even seeing something (e.g. Mapudungun pelelen ñi allfen ‘take a look at my wound’, lit. ‘you (sg) see my wound for me’). On the beneficiary’s side, possible interactions with particular patients or themes may be contextually negotiated; in some languages (Japanese and Korean), opening a window for someone may require a much more marked context than opening a door for a given benefactive construction to be felicitous. By a related token, benefaction can deviate from the prototype with respect to the animacy of either the agent or the beneficiary; benefactive external possession constructions may contrast felicitously with nonbenefactive internal possession constructions, for instance, when describing the handling of parts of inanimate wholes in such a way that the latter are beneficially affected as judged by a potential manipulator (e.g. Spanish benefactive le pinté la puerta al auto vs. nonbenefactive pinté la puerta del auto, both ‘I painted the car door’). The patient can range from concrete to abstract as well. A well-known alternative to the view proposed here is Shibatani’s (1996) influential “cognitive account” of such parametrizations as evidenced by the crosslinguistic variation of whether benefactive constructions are applicable to particular beneficial states of affairs or not. (It is important to note that Shibatani restricts the use of benefactive to those constructions where the beneficiary is an argument rather than an adjunct.) This author takes a ‘give’-schema as the point of departure for both the syntactic and semantic analysis of ‘give’-benefactives (e.g. Japanese boku-wa Hanako-ni hon-o katte yatta [1sg-top H.-dat book-acc buy.cvb give. pst] ‘I bought Hanako a book’), with “possessive control” at its conceptual center and the notion of beneficial effects as a “construal” intimately connected to it. Shibatani then suggests that progressive relaxations of the formal and functional restrictions imposed by the ‘give’-schema account for the absence or presence of



Benefaction proper and surrogation 117

translational equivalents via grammaticalized benefactives like the English socalled double-object construction (I bought Mary a book) in different languages. A continuum of situations like opening/closing the door for someone, sweeping the garden for someone, killing the centipede for someone, dancing for someone, and going to the market for someone, for example, is diversely treated in comparatively restrictive languages like English and German (where only the first, or the first two, are admissible), more liberal languages like Sinhala and Indonesian, and particularly liberal languages like Javanese, Chichewa, and Spanish (where many benefactives built on intransitives are admissible). Lastly, such a continuum can also have a diachronic application within one language to the extent that benefactive constructions are liberalized as the reflex of ‘give’ becomes more and more grammaticalized, i.e. semantically bleached. The view of benefaction advanced in the present article is basically a somewhat more detailed systematization of the semantic prototype and the deviations from it, and it is largely compatible with Shibatani’s view. Nevertheless, there are three important differences worth mentioning in the present context. First, benefactives as defined here cover all possible syntactic statuses of the beneficiary in a benefactive clause, from subject and object through to adjunct, or from privileged argument and minor argument through to peripheral nonargument. Second, the beneficial construal is seen as an integral part of the semantics of any benefactive construction. Third, the ‘on behalf of ’ reading is regarded as a notion related to benefaction but in principle independent from it, rather than considered something “[deriving] from the third semantic characteristics of the ‘give’ construction, [viz.] creating this favorable state on behalf of the possessor renders the giving situation highly desirable to the recipient” (Shibatani 1996: 181). The latter point is at the center of attention in the present article, and its exact contours shall become apparent in the next subsection. 3.2 Surrogation Substitution denotes states of affairs in which a given participant is replaced by a different participant, be it as a causing force (e.g. he killed the ambassador in the assassin’s place) or as an affected entity (e.g. we ate chicken instead of beef). There may indeed be a beneficial (or detrimental) effect on some participant, but it is crucial to observe that such a consequence is not conventionally denoted by a substitutive construction. In this respect, such constructions are quite unlike expressions signaling benefaction proper, be they specialized strategies like those including specific adpositions or broad ones like dative constructions or serial verb constructions based on a predicate of giving.

118 Fernando Zúñiga

Surrogation, therefore, can be seen as a link between substitution and benefaction proper: a given participant is beneficially affected by some other participant’s actions, but the resulting subevent is of a different nature than with benefaction proper. The beneficiary is portrayed as benefiting from the fact that s/ he does not have to perform a particular action thanks to the intervention of the surrogate (an agent or some other causer), without necessarily being in a position to interact better with any patient (like in object benefaction) or the action performed by the surrogate (like in event benefaction), and without indirectly benefiting from the action in an unspecified way (as in absolute benefaction). The fact that the beneficiary is relieved from a necessary and/or scheduled task is understood as beneficial in its own right, and portrayed as such. (This also answers the first and second questions mentioned in the beginning of Section 3: surrogation requires substitution and encompasses it, and surrogation is not merely an underlying meaning of several benefaction subtypes.) A special case of surrogation is found where the beneficiary is not in a position to carry out the action for a particular result (cf. Frawley’s example Dr. Frankenstein made his son a monster in the surrogacy reading). Consider Example (10) against this background. Finnish (Finno-Ugric, Finland) employs the allative case to signal benefaction proper and surrogation, as well as the ablative case for object benefaction and occasionally surrogation (not illustrated here). The prepositions takia ‘because’ and puolesta ‘instead’, however, allow only for a surrogation interpretation in addition to their basic meanings (reason and substitution respectively), not for an interpretation of benefaction proper (Kittilä 2010): (10) Finnish (Kittilä 2010: 251) a. Talonmies vaihto-i lampu-n takia-ni. janitor.nom change-3sg.pst lamp-acc because-1sg.psr ‘The janitor changed the lamp because of me.’ (allowed inference: ‘on my behalf ’, ‘so that I wouldn’t have to’) (excluded inferences: ‘in order to give it to me’, ‘in order for me to watch’) b. Talonmies vaihto-i lampu-n puolesta-ni. janitor.nom change-3sg.pst lamp-acc instead-1sg.psr ‘The janitor changed the lamp instead of me.’ (allowed inference: ‘on my behalf ’, ‘so that I wouldn’t have to’) (excluded inferences: ‘in order to give it to me’, ‘in order for me to watch’)

Similarly, the following English examples show that context — sometimes made explicit — can crucially affect the interpretation of a for-benefactive (underlined). In (11a), a patient/theme is manipulated via a transitive verb (note that permanent possession is not at stake), but in (11b) and (11c) it is an intransitive verb that serves as the base for the benefactive:



Benefaction proper and surrogation 119

(11) English a. Can you lift that for me? It’s too heavy and you are much stronger than I am. (Self-Journey.com, 15.07.10) b. Could you go shopping for me this afternoon? I really need to finish that term paper! (overheard on campus, University of Oregon, 12 June 2009) c. Rosario is so glad to see them. Before they got here, Rosario was a mess with everything that happened this morning. We went to the market for her, and she was so grateful that there are people here to help her out. Hopefully she’ll calm down now that her sister’s here and everything can calm down. (traveljournals.net, 24 July 2009)

In (11b), there is a semantic patient (whose exact description will probably be the topic of the subsequent exchange), and the person asking the favor is indeed the plausibly intended (ultimate) recipient of the purchased goods, but note that this is not asserted in the utterance. It is merely the most likely context, since the shopping could very well be for a charity, so that the ultimate recipient would be someone different from the beneficiary in the sentence. In (11c), the verb is more clearly intransitive, and the context is given in full in the original; the purchased goods are even more obliquely expressed — in fact, only implicated — than in (11b), and the intended reception of any items is only one possibility among many. (Rosario may be the town mayor, for instance, and she may simply be supposed to go to the market to bid the foreign vendors’ representative welcome there, or take a look at the public lighting on the square where the market takes place, without any goods being purchased for her.) By a similar token, consider the contrast between the two English sentences in (12). To adequately understand the most likely interpretations of these sentences, the reader probably needs to know that Peter Gabriel and Phil Collins were famous British pop singers active from, roughly, 1975 to 2005: (12) English (p.k.) a. Peter Gabriel sang for Phil Collins at the opening of the charity concert. b. Peter Gabriel sang for the audience one last time after the standing ovation.

Even though the grammar allows not only for a surrogation reading but also for an event benefaction reading of both sentences, the ‘on behalf of ’ reading is in all probability the preferred one for (12a) and the ‘for the enjoyment of ’ reading is certainly the default interpretation of (12b). Further observe that surrogation and event benefaction are not necessarily always possible as simultaneous readings of these utterances but can exclude each other. Peter Gabriel may have sung for Phil Collins — the expected performer — because the latter had lost his voice but was in the audience, or because Collins was stranded on the other side of the world at

120 Fernando Zúñiga .

the time of the concert (which was or was not recorded, so he may or may not have seen/heard/enjoyed Gabriel’s performance). Supposing that Gabriel habitually has audiences sing the chorus of one particular song towards the end of his concerts, if he sang on behalf of the audience, the latter presumably benefited from their not having to sing and from his singing. The default interpretation, however, is that the audience benefited from enjoying Gabriel’s performance only, because they were not actually supposed to sing after a standing ovation. Finally, note that, in Tashelhiyt (Afro-Asiatic, Morocco), the dative covers a wide variety of affected participants, including recipients and beneficiaries, like in the classical languages. In Example (13), the dative denotes a beneficiary in the context of a fantastic story, but surrogation is excluded as an admissible interpretation, because the magical cat was the only character capable of performing the action in question. Despite the fact that the beneficiary benefits from the cat’s excreting gold because he was not in a position to do so himself, this is not what the benefactive signals here; the dative exclusively expresses object benefaction proper: (13)

Tashelhiyt (Rapold 2010: 360) Ar a-s i-ttxxi umušš lli ddhb. ipfv dat-3sg 3sg.m-excrete.ipfv cat(m) ana gold ‘The cat excreted gold for him.’

3.3 Benefaction proper vs. surrogation Benefaction proper is seen here as including a resulting subevent that is semantically underspecified: it consists in the fact that the beneficiary can interact in some way with a patient/theme, or with the causing subevent as a whole, or derive some benefit indirectly from the state of affairs performed by the causer. The fact that the beneficiary cannot perform the action carried out by the causer is a possible additional inference in some contexts. It is this latter meaning that is the resulting subevent of surrogation, which does not entail any interaction with a patient/ theme or the causing subevent. These relationships are schematized from the perspective of the beneficiary in Table 3. The examples discussed in Subsection 3.2 showed how these two types of benefaction may interact in constructions that do not distinguish them formally. The next section focuses on constructions that do not mark beneficiaries of either type employing the same formal means.

Benefaction proper and surrogation 121



Table 3.  Benefaction proper vs. surrogation Resulting subevent

Inference (in some contexts)

Ia. Object benefaction

Benefit derived from possible interaction with patient of causing subevent

(Benefit derived from possibility of not performing the causing subevent)

Ib. Event benefaction

Benefit derived from possible inter­action with causing sub­event as a whole

I. Benefaction proper

Ic. Absolute benefaction Benefit indirectly derived from causing subevent Surrogation

Benefit derived from possibility of not performing the causing subevent

(Benefit derived from possible interaction with patient of causing subevent or subevent as a whole)

4. Surrogative strategies In light of the evidence presented in the last section, it is not possible to claim here that all benefactives are semantically/pragmatically specialized in such a way that surrogation is neither readily available as a reading when benefaction proper is excluded nor underlyingly present as a basic component of the construction’s meaning. Rather, I would like to argue in this section that some benefactives are specialized in revealing ways. The Spanish prepositional benefactives with por and para stand in complementary distribution (§4.1); the former expresses surrogation and the latter benefaction proper, with no overlapping or ambiguity/vagueness like the one found with English for. Toba has two benefactive applicatives -em and -ʔot (§4.2); while the former covers a range of different benefaction subtypes (sometimes including surrogation), the latter basically excludes surrogation (and might therefore be labeled “antisurrogative” benefactive). Lastly, the benefactive applicatives of Bemba work in such a way that surrogation is actually based on benefaction proper — not necessarily in semantic, but clearly in formal, terms (§4.3). 4.1 Spanish por and para The historical development of the Spanish markers por ‘for, by, through’ and para ‘for (the benefit of)’ is admittedly less than straightforward, since both descend from the result of the merger of Latin per and prō (Penny 2002: 243). Moreover, both elements cover a range of meanings, only some of which are directly related to benefaction. Nevertheless, to the best of my knowledge, the situation in all

122 Fernando Zúñiga

varieties of present-day Spanish is unambiguous: por denotes reason and surrogation, not benefaction proper, whereas para denotes purpose and benefaction proper, not surrogation. This distinction is clearly seen in clauses with an effected patient (14). Utterance (14a) is felicitous if s/he built the house in my stead, as a favor to me, or because of me in some other sense. By contrast, utterance (14b) is felicitous if s/he built the house intending me to inhabit it, either as a permanent resident or as a temporary/occasional lodger, or perhaps intending me to benefit from a built house in some other way. (14) Spanish (p.k.) a. Construyó la casa por mí. built.3sg art house for1 1sg.obl ‘S/he built the house for me.’ (‘because of me, for my sake, on my behalf ’) b. Construyó la casa para mí. built.3sg art house for2 1sg.obl ‘S/he built the house for me.’ (‘for me to inhabit, for me to enjoy it’)

If there is overlap in the meaning — i.e., if s/he built the house in my stead and with me as a future lodger/inhabitant in mind —, both expressions are available, depending on which side of the coin the speaker chooses to convey. Crucially, if there is no overlap in meaning, just one of the two possibilities is in order. The distinction is slightly different with intransitive verbs (15), but both prepositions are in complementary distribution in such cases as well. With the verb of artistic performance bailar ‘dance’, (15a) is felicitous if you were supposed to dance and s/he took your place, or if s/he did it because of you in sense that remains implicit. Utterance (15b) is felicitous only if you can actually “do something” with that performance, like watching or enjoying it: (15) Spanish (p.k.) a. Bailó por ti. danced.3sg for1 2sg.obl ‘S/he danced for you (sg).’ (‘because of you, for your sake, on your behalf ’) b. Bailó para ti. danced.3sg for2 2sg.obl ‘S/he danced for you (sg).’ (‘for you to watch, for you to enjoy it’)

If your dancing was out of the question, (15a) is ruled out; similarly, if you were not able to see or somehow else perceive the performance in some way, (15b) is not felicitous. To sum up: The use of para can conversationally implicate, but it does not imply, surrogation. By the same token, the use of por can conversationally implicate



Benefaction proper and surrogation 123

some sort of involvement on the part of the beneficiary, but it does not entail such a meaning. The semantic opposition is equipollent, unlike the one found in English between on behalf of, which is specialized, and for, which covers a wide range of senses (including ‘on behalf of ’) depending on a complex interplay of lexical and contextual factors. 4.2 Toba -em and -ʔot Toba (Guaycurú, Argentina) as analyzed by Censabella (2010) has a series of suffixes that applicativize predicates, one of which is apparently reserved for recipients and basically occurs only on the generic verb of giving -an. Two of the applicatives are especially relevant for our present purposes, viz. benefactive -em and locative -ʔot. The former typically introduces object beneficiaries, but also causes/reasons, maleficiaries, and surrogate beneficiaries (the latter usually with monovalent bases). In addition to its spatial function, the latter introduces beneficiaries that have some kind of authority, and the surrogative reading is excluded. Consider the examples with the (mono)transitive verb -men ‘sell’ in (16). In (16a), since the patient is explicitly identified as ‘my house’, surrogation is possible in principle but unlikely, and the affected participant introduced by the applicative is interpreted as a recipient. In (16b), since the nonpossessed patient does not impose the same restriction as in (16a), several readings are allowed, viz., reception, benefaction proper (probably absolute benefaction), and surrogation. By contrast, the affected participant in (16c) is a person with some authority — in the commercial exchange situation, the “boss” is identified with the customer/buyer — and the surrogative reading is ruled out. (16) Toba (Censabella 2010: 192) a. ʔAm se-men-em ñi i-maʔ. 2sg 1.act-sell-ben.appl deic 1sg.psr-home ‘I sell you (sg) my house.’ (you are my friend, not my boss; surrogation highly improbable) b. ʔAm se-men-em ñi nojik. 2sg 1.act-sell-ben.appl deic house ‘I sell you (sg) the house.’ / ‘I sell the house for you (sg).’ (you are my friend, not my boss; surrogation allowed) c. ʔAm se-men-ot a-so in-kanasto. 2sg 1.act-sell-loc.appl f-deic indet.psr-basket ‘I sell you (sg) that basket.’ (you are a customer; surrogation excluded)

124 Fernando Zúñiga

The examples in (17) illustrate a similar contrast with the verb -achaɢan ‘show’. Used without any applicative morphology, a matrix clause with this verb includes only the person doing the showing and the thing being showed (not illustrated here). Applicativization with the allative applicative -a licenses the experiencer without any expression of benefaction (17a). Benefactive applicativization accommodates an underspecified beneficiary — a participant advantageously affected via either benefaction proper or surrogation (17b). Lastly, (17c) excludes the surrogative reading and presents the beneficiary as someone with authority (in this context, someone judging the craftsman’s work): (17) Toba (Censabella 2010: 193) a. Ramaʒe ʔam j-achaɢan-a so l-ʔonatak. 3sg 2sg 3.act.tr-show-all.appl deic 3.psr-craftwork ‘He shows you (sg) his craftwork.’ (you can see his work; neither benefaction proper nor surrogation) b. Ramaʒe ʔam j-achaɢan-em so l-ʔonatak. 3sg 2sg 3.act.tr-show-ben.appl deic 3.psr-craftwork ‘He shows you (sg) his craftwork (for your benefit).’ (you will benefit from knowing about his craftwork; surrogation allowed) c. Ramaʒe ʔam j-achaɢan-ot so l-ʔonatak. 3sg 2sg 3.act.tr-show-loc.appl deic 3.psr-craftwork ‘He shows you (sg) his craftwork.’ (you are the craftwork teacher; surrogation excluded)

Example (17c) may raise the question whether the participant introduced by the locative applicative really is a beneficiary. I would like to claim here that this is the case. It is admittedly an unusual one, but not because it does not express the benefit’s nature in detail; prepositions like English for and Spanish para do not do that either, and run-of-the-mill datives can hardly be regarded as semantically specific in this respect. This beneficiary is unexpected from a Western European perspective because it is specialized for a feature related to the social standing of the participant (possibly in general, at least relative to the subject).4 In (18), with the intransitive verb -ʔonataɢan ‘work’, the rich man is portrayed as benefiting from the work done by the men in an unspecified way:

4.  Consider in this context the contrast between low-honorific and high-honorific verbs of giving in Japanese, which are employed in benefactive constructions specialized for different social scenarios. The similarities and differences between the Toba and the Japanese strategies are not pursued further here; cf. the revised edition of Zúñiga (2011a) (in preparation at the time of writing) for a detailed discussion.

Benefaction proper and surrogation 125



(18)

Toba (Censabella 2010: 192) Ra jale-pi ro-ʔonataɢan-ot seʔeso saλaɢanek. dem man-coll 3.act.itr-work-loc.appl dem.pron rich.man ‘The men work for that rich man.’

While intransitive verbs like -ʔonataɢan ‘work’ may favor the surrogative reading with the benefactive applicative -em, those like the performance verb -ʔon ‘sing’ allow interpretations in terms of both event benefaction and surrogation (19a) but show no discernible effects on the acceptability of the surrogative reading with the locative applicative -ʔot (19b): (19) Toba (Censabella 2010: 194) a. Ramaʒe ro-ʔon-aɢaa-teg-em so l-apiʔ. 3sg 3.act.itr-sing-prog-antipass-ben.appl deic 3.psr-grandfather ‘He is singing for his grandfather.’ (both readings are possible: ‘for him to hear/watch’ and ‘on his behalf ’) b. Ramaʒe ro-ʔon-aɢaa-teg-toʔot ñi Dios. 3sg 3.act.itr-sing-prog-antipass-loc.appl deic God ‘He is singing for God.’ (surrogation excluded)

To sum up: The applicative -em introduces an affected participant to the clause, which is interpreted as a recipient, a reason, a beneficiary, or a maleficiary depending on the specific verb on which it occurs, other semantic elements present in the clause, and the context of utterance. The applicative -ʔot introduces locations to the clause, as well as beneficiaries that are unspecified as to the exact nature of the benefit but are unmistakably identified as not benefiting from surrogation; instead, such participants are understood as persons having some kind of authority. 4.3 Bemba applicatives The literature on Bantu applicatives has shown that, in many languages of this family, applicatives usually license a wide semantic range of possible applied participants, viz. different subtypes of beneficiaries, maleficiaries, locations, directions, reasons, and instruments (cf. e.g. Peterson 2007 and the references therein). Typically, disambiguation among all these meanings in applicativized clauses is achieved simply by resorting to lexical information and/or knowledge about the context of utterance — not unlike the situation found in the interpretation of Romance or Slavic dative-marked participants. In Bemba (Bantu M42; Zambia) as described by Marten & Kula (2011), however, for the applicativized construction to allow a surrogative reading, additional morphology is needed: the verbal enclitic =kó is obligatory for the expression of

126 Fernando Zúñiga

such a meaning (20b). Note that the ‘instead of ’ meaning is instantiated by the applicative plus the enclitic when the applied object is the idle agent, whereas it is expressed by the non-applicative predicate and a different construction based on the preposition úkúcila ‘instead of ’ when a given referent would have been patientive had it been centrally involved in the relevant state of affairs (20c). (In the latter case, no part of the meaning appears to be related to benefaction; the yield is merely substitutive.) (20) Bemba (Sadler 1964: 271–272, quote in Marten & Kula 2011: 13) a. N-ka-samb-il-a Chali úmu-ána. 1sg.subj-fut-wash-appl-fv Ch. I-child ‘I will bathe the child for Chali.’ b. N-ka-samb-il-a=kó Chali úmu-ána. 1sg.subj-fut-wash-appl=loc2 Ch. I-child ‘I will bathe the child instead of Chali (bathing him).’ c. N-ka-samb-a úmu-ána úkú-cila Chali. 1sg.subj-fut-wash-fv I-child XV-surpass Ch. ‘I will bathe the child instead of (bathing) Chali.’

The verbal enclitic =kó is transparently related to a locative demonstrative element belonging to a series usually identified by their noun class correspondence as XVI (pV, specific location), XVII (kV, approximate location or direction), and XVIII (mV, interior location), respectively. The second element of these series (kV) is the one relevant for our present purposes; it appears both as a class prefix (ku‑) and in the anaphoric demonstrative ukó in a non-predicative unit like (21a), but also as a purely locative deictic =kó in a predication like (21b): (21) Bemba (Marten & Kula 2011: 18) a. ku-ngáanda ukó XVII‑IX.house dem.XVII ‘by/to that house’ b. N-ka-y-a=kó maílo. 1sg.subj-fut-go-fv=loc2 tomorrow ‘I will go there tomorrow.’

Three characteristics of =kó are particularly interesting in the present context. First, it is the only locative enclitic that denotes surrogation; applicativized verbs with =pó ‘locative1’ and =mó ‘locative3’ do not receive an interpretation different from what the simple application of the compositionality principle dictates, i.e. a locative one. Second, its function seems to be best understood as marking what Marten & Kula (2011) call “abstract space of agency.” This can be seen from two facts: its interpretation is not restricted to substitutive agency (even though



Benefaction proper and surrogation 127

this is arguably the preferred choice), but a more “indirect” relation is also possible with some verbs and in some contexts (22a) (i.e. absolute benefaction); moreover, use of the enclitic with a non-applicativized verb can denote counter-expectation (22b) (i.e., the children are not the usual cooks, or do not usually cook potatoes), which in turn might be considered the link to the applicativized version in (22c), where the expectation is made explicit (i.e. s/he is the habitual cook, or the one who would have been expected to cook potatoes): (22) Bemba (Marten & Kula 2011: 21–22) a. A-alí-n-sós-éel-e=kó. I.subj-rem.pst-1sg.obj-speak-appl-fv=loc2 ‘He spoke in my favor/defense.’ b. Ábá-na bá-á-!ípík-a=kó ífy-umbu. II-child II.subj-pst-cook-fv=loc2 VIII-potato ‘The children have (surprisingly) cooked potatoes.’ c. Ábá-na bá-á-mw-ípík-il-a=kó ífy-umbu. II-child II.subj-pst-III.obj-cook-appl-fv=loc2 VIII-potato ‘The children have cooked potatoes on his/her behalf.’

Third, with some verbs of motion like -pílíbuka ‘turn’, -bútúka ‘run’, or -tólóka ‘jump’, =kó encliticized to an applicativized verb denotes surrogation (23c) although the applicativized verb on its own does not allow an interpretation in terms of benefaction proper (23b): (23) Bemba (Marten & Kula 2011: 23) a. Ábá-icé bá-lée-tólok-a bá-mayó. II-child II.subj-prog-jump.over-fv II.mother ‘The children are jumping over the mother.’ b. Ábá-icé bá-lée-tólok-el-a bá-mayó. II-child II.subj-prog-jump.over-appl-fv II.mother ‘The children are jumping onto the mother.’ (not: ‘The children are jumping for the mother.’) c. Ábá-icé bá-lée-tólok-el-a=kó bá-mayó. II-child II.subj-prog-jump.over-appl-fv=loc2 II.mother ‘The children are jumping for / on behalf of the mother.’ (not: ‘The children are jumping onto the mother.’)

If Marten & Kula’s analysis is correct, surrogation in Bemba is neither a basic meaning underlying any benefactive construction nor a general default meaning to which benefactive constructions can resort whenever a more concrete interpretation related to reception(-benefaction) is not (readily) available. This Bantu language has developed a marker that is not quite specialized for the expression

128 Fernando Zúñiga

of surrogation but covers a fairly coherent and abstract range of related meanings, one of which is surrogation. 5. Conclusions The view presented here is not a radical departure from previous accounts of the complex of thematic (sub)relations orbiting the notion of beneficiary. Thanks to a definition of benefaction proper in terms of prototype and periphery, on the one hand, and an account of surrogation as a type of benefaction in its own right (rather than a basic notion underlying other benefaction types, e.g.), on the other, descriptivists and typologists are not forced to account for benefactives in a way that may be too restrictive, assuming an a priori connection between them. Even though surrogation can be, and often is, present as a secondary reading of benefaction proper, depending on lexical and contextual factors, it need not be, as widerange benefactives like English for show, and some strategies of languages like Toba may actually exclude the former interpretation altogether. Furthermore, benefactive strategies like the ones found in Spanish and Bemba show that both subtypes of benefaction can have specialized means of expression — formally equipollent in the former case and privative in the latter, where the surrogative construction is less basic than the benefactive proper one. In short, and even though the eventual treatment given to these notions ultimately depends on the particular theory of thematic relations an analyst or theoretician ends up endorsing, the account presented here argues in favor of regarding benefaction proper and surrogation as related to intimately connected but different semantic roles, possibly along lines similar to possessors and recipients. What does the above mean for Kittilä’s (2005) typology of benefaction? With the alternative definitions proposed here of the categories involved, the question of function/form correspondences becomes: Are recipients formally marked like any kind of beneficiaries or not, and are different subtypes of beneficiaries marked alike or differently? The prototype account of benefaction proper predicts that object benefaction, event benefaction, and absolute benefaction constitute a natural progression, which in turn means that formal expression should be found to follow that path — an empirical claim worth pursuing further, of course. The benefaction proper vs. surrogation account predicts that any logical combination between formal patterns corresponding to these two benefaction types, as well as any pairings with the expression of reception, can be found in the languages of the world. In this respect, the present study should be regarded as complementary to Kittilä’s in that the available evidence suggests that all possible combinations do in fact exist; further research shall elucidate the question of which combinations



Benefaction proper and surrogation 129

are more frequent and/or stable, and why. Finally, given the intimate connection between reception and benefaction proper with notions like possession, goal of motion, cause/reason, and purpose, a revision of Kittilä’s typology would benefit from a systematic crosslinguistic investigation of these domains as well. The Finnish and Spanish data suggest an interesting connection between surrogation and cause/reason (bypassing, or excluding, benefaction proper), which is worth taking seriously and exploring further when attempting to draw elaborate semantic maps covering all these notions.5 Related to this, a comparison between the further evolution of three different locative strategies, viz. English on, Toba -ʔot, and Bemba =kó, shows that such elements can follow rather different development paths related to benefaction. English on has acquired a malefactive yield in utterances like it rained on us (like the Gumer third object agreement set, cf. Völlmin 2010), while the markers in Toba and Bemba have turned into benefactives instead, but fairly dissimilar ones: Toba -ʔot rules surrogation out while Bemba =kó is actually specialized for that subtype. Authors of descriptive and/or diachronic studies have to keep in mind that locatives can turn into such different things in the course of their evolution. By contrast, benefactives based on a verb of giving, like serial verb constructions or morphological applicatives that can be traced back to such a verb, seem to follow the path from object benefaction through event benefaction to absolute benefaction, at some point possibly allowing for an extension into the domain of surrogation for some constructions. That this need not be the case can be seen from the so-called double-object construction in English and varieties of Dutch, for example, which covers a subset of object and event benefaction and displays an interesting range of variation with respect to its productivity. The present article is not a systematic study of the development paths of locatives and verbs of giving in the context of benefaction, however; such a study remains to be undertaken. What the present article has tried to show is that conceiving of surrogation as a subtype of benefaction in its own right along the lines proposed here is a useful analytical move. Further in-depth studies of benefactives in general and surrogatives in particular shall provide additional evidence for the approach presented here and/or help refine it.

5.  To be sure, the semantic connection between recipients, goals, and beneficiaries has been noted in innumerable studies. Studies particularly relevant for typologists in the context outline here would be those close in spirit to Fillmore (2004) and in scope to Schmidtke-Bode (2010). Rice & Kabata’s (2007) study of allatives deserves special mention in this context.

130 Fernando Zúñiga

Acknowledgements I am indebted to several persons who commented on an earlier version of this chapter. John Newman gave me some valuable advice as to how to improve the chapter’s readability; Ekkehard König helped me strengthen my case; and Daniel Hole motivated me to rethink the whole thing.

Abbreviations acc accusative, act actor, agt agent, all allative, ana anaphoric, antipass antipassive, appl applicative, art article, ben beneficiary, coll collective, cvb converb, dat dative, deic deictic, dem demonstrative, doc double-object construction, f feminine, fut future, fv final vowel, indet indeterminate, ipfv imperfective, itr intransitive, loc locative, m masculine, nom nominative, obj object, obl oblique, prog progressive, pron pronoun, psr possessor, pst past, rec recipient, rem remote, sg singular, subj subject, top topic, tr transitive 1, 2, 3 refer to grammatical persons I, II, III, etc. refer to noun classes

References Blake, Barry. 1994. Case. Cambridge: Cambridge University Press. Bohnemeyer, Jürgen, Nicholas Enfield, James Essegbey, Iraide Ibarretxe-Antuñano, Sotaro Kita, Friederike Lüpke & Felix Ameka. 2007. Principles of event segmentation in language: The case of motion events. Language 83(3). 495–532.  doi: 10.1353/lan.2007.0116 Bohnemeyer, Jürgen & Eric Pederson (eds.). 2011. Event representation in language and cognition. Cambridge: Cambridge University Press. Censabella, Marisa. 2010. Beneficiaries and recipients in Toba (Guaycurú). In Fernando Zúñiga & Seppo Kittilä (eds.), Benefactives and malefactives: Typological perspectives and case studies, 185–201. Amsterdam: John Benjamins.  doi: 10.1075/tsl.92.07cen Fillmore, Charles. 2004. Circumstance concepts. In G. Booij, et al. (eds.), Morphology: An international handbook on inflection and word-formation, Vol. 2, 1117–1130. Berlin: de Gruyter. Frawley, William. 1992. Linguistic semantics. Hillsdale, NJ: Lawrence Erlbaum Associates. Hole, Daniel. 2008. Dativ, Bindung und Diathese. Berlin: Humboldt University Habilitation thesis. Kittilä, Seppo. 2005. Recipient-prominence vs. beneficiary-prominence. Linguistic Typology 9(2). 269–297. Kittilä, Seppo. 2010. Beneficiary coding in Finnish. In Fernando Zúñiga & Seppo Kittilä (eds.), Benefactives and malefactives: Typological perspectives and case studies, 245–270. Amsterdam: John Benjamins.  doi: 10.1075/tsl.92.10kit Marten, Lutz & Nancy Kula. 2011. Benefactive and substitutive applicatives in Bemba. Ms., SOAS and University of Essex. Palmer, Frank R. 1994. Grammatical roles and relations. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9781139166638

Benefaction proper and surrogation 131



Penny, Ralph. 2002. A history of the Spanish language. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9780511992827 Peterson, David. 2007. Applicative constructions. Oxford: Oxford University Press. Platt, J. 1971. Grammatical form and grammatical meaning: A tagmemic view of Fillmore’s deep structure case concepts. Amsterdam: North-Holland. Rapold, Christian. 2010. Beneficiary and other roles of the dative in Tashelhiyt. In Fernando Zúñiga & Seppo Kittilä (eds.), Benefactives and malefactives: Typological perspectives and case studies, 351–376. Amsterdam: John Benjamins.  doi: 10.1075/tsl.92.15rap Rice, Sally & Kaori Kabata. 2007. Crosslinguistic grammaticalization patterns of the allative. Linguistic Typology 11(3). 451–514.  doi: 10.1515/LINGTY.2007.031 Sadler, Wesley. 1964. Untangled Bemba: A language of Northern Rhodesia, Central Africa. Kitwe: The United Church of Central Africa in Rhodesia. Schmidtke-Bode, Karsten. 2010. The roles of benefactives and related notions in the typology of purpose clauses. In Fernando Zúñiga & Seppo Kittilä (eds.), Benefactives and malefactives: Typological perspectives and case studies, 121–146. Amsterdam: John Benjamins.  doi: 10.1075/tsl.92.05sch

Shibatani, Masayoshi. 1996. Applicatives and benefactives: A cognitive account. In M. Shibatani & S. Thompson (eds.), Grammatical constructions: Their form and meaning, 157–194. Oxford: Clarendon Press. Smith, Tomoko. 2005. Affectedness constructions: How languages indicate positive and negative events. Berkeley: University of California at Berkeley dissertation. Somers, Harold. 1987. Valency and case in computational linguistics. Edinburgh: Edinburgh University Press. Song, Jae Jung. 2010. Korean benefactive particles and their meanings. In Fernando Zúñiga & Seppo Kittilä (eds.), Benefactives and malefactives: Typological perspectives and case studies, 393–418. Amsterdam: John Benjamins.  doi: 10.1075/tsl.92.17son Teng, S.-H. 1975. A semantic study of transitivity in Chinese. Berkeley: University of California Press. Van Valin, Robert & Randy LaPolla. 1997. Syntax. Structure, Meaning and Function. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9781139166799 Völlmin, Sascha. 2010. Benefactives and malefactives in Gumer (Gurage). In Fernando Zúñiga & Seppo Kittilä (eds.), Benefactives and malefactives: Typological perspectives and case studies, 317–330. Amsterdam: John Benjamins.  doi: 10.1075/tsl.92.13vol Zúñiga, Fernando. 2011a. The grammar of benefaction: A crosslinguistic study. Zurich: University of Zurich Habilitation thesis. Zúñiga, Fernando. 2011b. Why should beneficiaries be subjects (or objects)? In S. Kittilä, K. Västi, and J. Ylikoski (eds.), Case, Animacy and Semantic Roles, 329–348. Amsterdam: John Benjamins.  doi: 10.1075/tsl.99.12zun

Exploring a diachronic (re)cycle of roles The Dative complex from Latin to Romance Chiara Fedriani and Michele Prandi

University of Bergamo / University of Genoa

In this chapter we explore the struggle between the use of the Dative case and the competing strategy featuring the preposition ad ‘to’ and the Accusative from Latin to Early Romance. Unlike the Dative, the prepositional strategy is semantically transparent, since ad ‘to’ has a clear allative meaning. We first consider the diachronic development of the roles involved in the Dative-marked complex within a chronological span ranging from Plautus to the Vulgate and show that competing manifestations featuring ad are conditioned by semantic factors, since the extension of the prepositional strategy can be better explained in terms of metaphoric and metonymic processes. We discuss the gradual expansion of the prepositional turn in Early Romance with a view to exploring the paths along which it gradually took over the functions traditionally associated with the Dative complex. Building on these data, the chapter assesses the theoretical implications for a better understanding of competing multifunctional devices that encode role complexes from a diachronic perspective and shows how a pool of synchronic variation can trigger and constrain linguistic change. Keywords: dative, grammaticalization, role complex, Latin, Romance languages

1. Introduction In Classical Latin, the Dative is a semantically opaque grammatical case that encodes a composite constellation of grammatical relations and roles, including indirect object, the second argument of some two-place verbs, referred to by many linguists as Dative object (see, e.g., Levin 1993), and some free adjuncts, as we shall see in the course of this chapter. Already in Early Latin, however, the Dative function was occasionally expressed by means of a competing strategy featuring the preposition ad ‘to’ and the Accusative. Even as early as in Plautus (c. 254– 184 BCE) the expression of the third argument of some verbs alternates between Dative (ex. 1a) and the competing prepositional form (ex. 1b): doi 10.1075/bct.88.06fed 2016 © John Benjamins Publishing Company

134 Chiara Fedriani and Michele Prandi

(1) a. b.

Ego illis captivis aliis 1SG.NOM that.DAT.PL prisoner.DAT.PL other.DAT.PL documentum dabo example.ACC.SG give.FUT.1SG ‘I shall give an example to those other prisoners’ (Plaut. Capt. 752) Ego hunc… ad carnificem dabo 1SG.NOM this.ACC.SG to executioner.ACC give.FUT.1SG ‘I shall give him to the executioner’ (Plaut. Capt. 1019)

Unlike the Dative, the prepositional strategy is semantically motivated and transparent. The form ad + Accusative, where the preposition ad ‘to’ has an allative meaning, is borrowed from the expression of the Goal in the presence of verbs of motion and displacement due to a metaphorical transfer: the third arguments of some verbs that take the Dative — in particular the Addressee of an act of communication and the Recipient of a transfer — are reinterpreted as if they were the Goal of a motion. In Late Latin, the substitution became very general throughout most of the Empire (see Bastardas Parera 1953: 44–47, Löfstedt 1961: 221ff., Väänänen 1963: 120 for an overview), slowly encroaching on the domain of the Dative and ultimately taking over all of its uses in Romance languages (cf., e.g., Maiden 1996, Zamboni 2000). In modern Romance languages, the same form a/à ‘to’ + NP, which had begun its career as a motivated and transparent expression of the Goal, has become as opaque as the Dative case in Latin. This is the essence of the historical cycle that this chapter seeks to describe. A highly grammaticalized and semantically opaque form — the Dative case — is threatened by a competing transparent form whose spread is motivated by a chain of metaphorical extensions. Once the innovative form has won its battle, at the moment where the case system collapses during the transition from Latin to the Romance languages, the transparent form undergoes a process of grammaticalization that changes it into an opaque form of expression that conveys a constellation of grammatical relations and roles almost as wide and heterogeneous as that covered by the Dative case in Latin. In order to fully appreciate the radical nature of the historical cycle under scrutiny, it should be stressed that the Latin Dative grammaticalized as a multifunctional case which is not always bound up with the idea of motion towards a goal,1 1.  Although it has been noted that the Latin Dative seems to indicate the “pole of orientation” toward which a process or relation is projected, scholars also firmly maintain that the Dative was not properly felt in Latin to be the dedicated case to express spatial relations (see, e.g., Van Hoecke 1996: 19). Evidence comes for instance from the fact that one cannot find prepositions taking the Dative to convey Goal or Source and that a number of Prepositional Phrases, and not the use of the bare Dative, were preferred for this purpose (on the “locative” origin



Exploring a diachronic (re)cycle of roles 135

although compatible with it in some of its uses, whereas the innovative form is driven by precisely this feature and entails the semantic component of displacement to a destination. According to this premise, the gradual expansion of the innovative form in the functional domain of the Dative cannot be random, but is expected to follow a well-defined, revealing path of decreasing semantic motivation. More precisely, we will show that the expansion of the prepositional strategy at the expense of the Dative case has been triggered and constrained by metaphorical transfers supported by metonymic shifts, and is therefore deeply rooted in the specific semantics of different roles occurring in different contexts. As a matter of fact, the data show that this expansion begins with verbs that frame processes where motion is either really involved — for instance, mittere ‘to send’ — or metonymically connected in a direct way — for instance, scribere ‘to write’ in its communicative sense — and is subsequently extended to actions like dicere ‘to say’ that can be seen as kinds of motion in a metaphorical sense only. In the former cases, the role of metaphor is restricted to the categorization of the Addressee or Recipient as a kind of Goal. In the latter, the whole transaction is reshaped after the model of motion. The key point of this historical evolution is that the drift phases in Latin were transparent semantic motivation phases, leaving to the Dative all the roles that are not compatible with the scheme of a motion towards a Goal. This implies that even in Late Latin the prepositional form has not yet lost its semantic motivation. The drift is accomplished in the Romance languages, where the loss of the Dative case leads to the extension of the prepositional form to the whole constellation of the Dative’s former uses — namely, to the non-allative instances of indirect object and to the whole range of adjuncts. This move requires as a preliminary condition the total loss of the semantic motivation that has triggered the whole drift, that is, the total loss of the association between the form ad/a/à ‘to’ + NP and the Goal of a motion. The outcome is that the Romance constellation of grammatical relations and roles expressed by the prepositional form includes the entire heritage of Latin Dative and even more. The aim of this chapter is both to provide empirical support to the hypothesis concerning the diachronic cycle sketched above and to open up a critical discussion about its theoretical background.2 At the empirical level, we will try to unravel the micro-dynamics underlying the shift from the Dative to the prepositional of the Latin Dative, however, see Bennett 1914: 101–103, Villar 1981; see Hofmann & Szantyr 1972[1965]: 86–87 for a review of the discussion). 2.  Although valuable observations have been made about the competition of the Dative vs. the prepositional strategy (cf., e.g., Ernout & Thomas 1953: 69–70, Kühner & Stegmann 1976: I.519, Théoret 1982, Selig 1991: 198, Van Hoecke 1996, Baños Baños 1996, 2000, 2009, Christol 1998: 482, Luraghi 2010, Adams 2011: 266–268), these observations are confined to the dis-

136 Chiara Fedriani and Michele Prandi

form in its different stages. We will take into account the historical development of the Dative constellation based on the uses of a significant sample of verbs documented in the chronological span ranging from Plautus (254–184 BCE) to the Vulgate (5th century CE) and finally to the Romance languages. At the theoretical level, the most interesting question has to do with the different coding regimes that coexist in language structures and take turns in diachronic changes, governing in different ways the relationship between forms of expression, grammatical relations, and conceptual relations. The remainder of this chapter is structured as follows. After making explicit some terminological points, we go on to introduce the theoretical background in more detail in Section 2. We will then describe the main functions of the Latin Dative, discussing the entire constellation of roles encoded by the Dative case in Latin (Section 3). In the second part of the chapter we will consider the diachronic development of the roles involved in the Dative-marked complex and show that competing manifestations of these roles, notably the Prepositional Phrase featuring ad + Accusative, are conditioned by semantic factors, in the sense that the extension of the competing strategy can be better explained in terms of metaphoric and metonymic verb-specific processes (Section 4). In Section 5 we explore the gradual expansion of the prepositional turn in some Romance languages, where case distinctions were lost for nouns and partially preserved in pronouns only. Section 6 reviews the findings, and, building on data from Latin, assesses the theoretical implications for a better understanding of competing multifunctional devices that encode role complexes from a diachronic perspective. Before turning to the data, it seems useful to briefly make explicit some terminological issues that are relevant to a better understanding of the following discussion. The distinction between arguments and non-argumental roles, however problematic in some points of detail, will be taken for granted. Non-argumental roles that enrich the core of a process, currently named adjuncts, will be referred to as margins,3 following Longacre (2007[1985]). Among the arguments relevant to the present discussion, Goal refers to the spatial relation that identifies the end point of a motion or a displacement. Addressee refers to the third argument of an act of communication, while Recipient denotes cussion of context-specific or verb-specific alternations and do not amount to a complete, diachronically satisfying account. 3.  The advantage of the term margins is that it leaves open the question about the different positions taken by non-argumental roles within the structure of an extended process. When it is used alone, margin simply means non-argumental role. When saturated by a complement, the term has a more exact reference to different layers within a complex hierarchy: to margins of the process, of the predicate, of the main verb, of a noun (Prandi 2004: 268–280).



Exploring a diachronic (re)cycle of roles 137

the third argument of an act of giving. The relationship between the Addressee and a speech act is the same as that between a Recipient and an act of giving. For the sake of our argument, the relevant point is that both roles can be seen as the end points of a transaction and therefore be metaphorically categorized as kinds of Goal. Thanks to this analogy, the two roles behave in the same way, as will become manifest in the course of our analysis; in particular, the shift from Dative to the prepositional form takes place exactly under the same conceptual conditions for Recipient and Addressee. The case of Beneficiary is more complex and disputable. The idea of Beneficiary is relevant at two logically distinct levels. Within the valency scheme of three-place verbs, Beneficiary is not an argument at the same level as Recipient and Addressee but an implication of either, along with Maleficiary. The Recipient of a cake and the Addressee of a compliment, for instance, enjoy a benefit in the same obvious way that the Recipient of a poisoned apple and the Addressee of an insult are damaged. Whereas the conceptual profile of either Addresse or Recipient qualifies the process that contains it, such positive or negative implications do not. As an autonomous role, Beneficiary is not an argument of a specific class of verbs but a peripheral role compatible with any kind of action. Any saturated action, from cutting wood to running home, can be performed for or against someone. The set of peripheral roles involved in our discussion include, beside Beneficiary and Maleficiary, External Possessor (König & Haspelmath 1998), Involuntary Causer, termed Involuntary Agent by some authors (cf. Haspelmath 1993: 291–3, Kittilä 2005, Næss 2007: 82) and Substitute for the Agent, i.e. someone who, from the perspective of the speaker, could or should have performed the action instead of the actual agent. All these peripheral roles are relevant in that they can be entrusted to the form a/à + NP in the Romance languages. The label Role complex refers to the constellation of roles associated with one form of expression. Dative role complex, in particular, refers to the whole set of roles associated with the Dative case. Such clusters of roles, which are cross-linguistically common,4 constitute the other side of the coin with regard to the phenomenon of Role accumulation (Lehmann 2006), which takes shape when a single form within a single construction expresses more than one role at a time. An example of role accumulation is the Italian Prepositional Phrase a Paolo ‘to Paul’ in a sentence like Maria ha lavato i vestiti a Paolo ‘Mary washed the clothes to Paul’, where the Prepositional Phrase a Paolo cumulates the roles of Beneficiary (Mary did it for Paul), External Possessor of the clothes (Mary washed Paul’s clothes), and Substitute for the Agent (Mary did it instead of Paul). 4.  A clear example of role clustering within a Role complex is constituted by the English indirect object to + NP, which is able to express, under different conditions, such roles as Goal (go to London), Recipient (give a book to Mary), and Addressee (tell a story to Mary).

138 Chiara Fedriani and Michele Prandi

2. Theoretical background: Relational coding and punctual coding It is frequently observed that the relation between roles, i.e. conceptual relations, and linguistic forms of expression, is not synchronically uniform while diachronically changing. In the current literature, such oppositions as motivated vs. arbitrary coding, transparent vs. opaque expression, grammatical vs. notional or semantic case (see, e.g., Kuryłowicz 1964: Ch. VIII, Blake 2001[1994]: 31ff.), along with such categories as semantic bleaching, raise precisely this question. In our opinion, the coexistence of opposite ways of coding is inadequately accounted for as long as coding is seen as an unidirectional vector, either from form to content, where grammatical relations are assumed to shape conceptual relations, as in formal approaches, or from content to form, where forms of expression are assumed to mirror conceptual relations, as in functional and cognitive approaches.5Linguistic data challenge both these unidirectional conceptions of coding: in particular, opaque and highly grammaticalized forms of coding challenge the idea of a logical priority of conceptual relations, whereas motivated and transparent forms of coding call into question the idea of a logical priority of autonomous linguistic forms. The coexistence of coding regimes taking opposite directions is best accounted for if we reject the sharp alternative and the underlying presupposition of the unidirectionality of coding. Rather, the coding vector should be conceived of as bidirectional: two opposite and complementary ways of coding, i.e. from form to content and from content to form, interact within the structure of each sentence. Each sentence has a core formed by a network of grammatical relations whose formal structure, like the structural skeleton of a building, is independent of its changing conceptual contents and social functions. Grammatical relations like subject and direct object are not defined in isolation but on the basis of their relation with the structure of the sentence as a whole, and therefore as relations belonging to a unitary network. This is the reason why we speak of relational coding. 5.  The formal stance is given its most radical formulation by Chomsky: “grammar is autonomous and independent of meaning” (1957: 17), and “uniquely determines […] semantic interpretation” (Chomsky 1966: 5). The functional and cognitive approaches share the idea that syntax is both instrumental and iconic: see, for instance, Dik (1989(1997: 8)): “Semantics is regarded as instrumental with respect to pragmatics, and syntax as instrumental with respect to semantics. In this view there is no room for something like an ‘autonomous’ syntax”. However, functional linguists tend to adopt a moderate version of iconicity — the meaning of a sentence mirrors an independent conceptual structure — whereas cognitive linguists lean towards a more radical stance: the meaning of a sentence depicts both the content of a communicative intention and a contingent experiential situation. According to Haiman (1985: 2), for instance, “the linguistic form is a diagram of conceptual structure”. According to Langacker (1992[1991]: 35), the meaning of a sentence is the “image” of “a particular event known in full detail”.



Exploring a diachronic (re)cycle of roles 139

Grammatical relations are independent of any content of role and are able to host different roles compatible with different main predicative terms — typically, verbs. Grammatical relations are constant, whereas roles are variable. To take an example that is crucial to the present context, the Dative case in Latin comes to express such heterogeneous arguments as the Addressee of dico ‘I say’, the Recipient of do ‘I give’, the second argument of irascor — irascor alicui, ‘I become angry with someone’ — and the third argument of nego ‘I deny’, precisely because it is not bound to any in particular.6 In all of these argument structures, the use of the Dative case does not mirror semantic configurations of participant roles, but relies instead on the grammatical relational properties of syntactic structures considered as pure schematic forms that are independent of the content they convey. As a result, semantic features depending on different verbal semantics are levelled out by the networks established by grammatical relations (cf. Lehmann 1991: 235). A sentence core is surrounded by different layers of margins. Among margins, coding runs from concepts to forms, using individual linguistic forms to mark individual conceptual relations that are logically prior and therefore accessible to consistent thought — to inference — independent of the form of expression. This means that among peripheral roles the structure takes shape at the conceptual level, while each form of expression becomes part of a structure only in that it performs the instrumental function of expressing a given role. If an expression, in turn, succeeds in expressing a given role it is not as a node of a relational structure as a whole, but in isolation, owing to its inner properties, and in particular thanks to the content of the preposition as far as coding is concerned and, if necessary, in view of its conceptual content as far as inference is concerned. An instrument, for instance, is defined with reference to the conceptual structure of an intentional action performed by a free and responsible agent. Against this premise, the standard form of expression of Instrument in English — with + NP — succeeds in expressing this role due to the synergy between the meaning of the preposition — which encodes nothing other than a generic subordinative co-occurrence — and inference: given John felled the tree with an axe, for instance, it is consistent to infer that the axe is there as an instrument against the background of the complex conceptual structure. These are the reasons why we speak of a punctual form of coding. Insofar as grammar is concerned, the forms of expression of margins are annexed to the core of the process independently of one another and owing to a choice made by the speaker and governed by conceptual consistency. Given such an action as John felled the tree, for instance, the prepositional phrase with + NP 6.  Note that nego has a number of different, but related meanings, the most frequent being ‘I say no’. For the sake of the present study, however, we only consider those cases in which the verbs mean ‘I deny (a thing to someone)’.

140 Chiara Fedriani and Michele Prandi

does not enter into relation with the core of the sentence thanks to its formal properties, but only in that it has been interpreted as the expression of a given role: it expresses the Instrument in John felled the tree with an axe, the Co-performer in John felled the tree with Peter, Manner in John felled the tree with rage, and many other consistent roles, as in John walked home with an axe; John left the room with a smile.7 The distinction between relational and punctual coding is relevant to iconism and conceptual motivation. A relationally coding expression is barred to conceptual motivation, as it is able to be filled by any role controlled by the main verb. The most striking example is grammatical subject, which can host almost any role whatsoever, but the case of indirect object is also significant in this regard. A punctually coding expression is iconic on two levels. It is iconic in a global sense, that is, diagrammatic (Jakobson 1966), since a given phrase is integrated into the structure of a complex expression insofar as it identifies a role belonging to an independent conceptual structure; the structure of the expression mirrors the structure of an independent complex concept. Moreover, it is iconic in a local sense in that the form of expression of a given role is motivated by its content. On the one hand, the preposition contributes with its content to identifying the relevant conceptual relations: the Latin preposition ad ‘to’, for instance, contributes with its allative content toward identifying the goal of a motion. On the other hand, when the content of the preposition is not rich enough to identify a consistent conceptual relation, as in the case of with, its function is taken over by inference, which is motivated in that it relies on the connected conceptual content. The two coding regimes are not statically in opposition to each other, but coexist in a dynamic form of equilibrium that is threatened at any moment. The reason lies in a significant gap between the formal and functional structure of meaningful sentences: the functional core formed by the verb and its argument does not necessarily coincide with the formal core shaped by a network of grammatical relations, since certain arguments of some verbs are encoded in a punctual way, that is, in the same way as margins. This point is particularly relevant to our discussion, as it shows that the two coding regimes can enter into competition

7.  The range of roles expressed by the preposition cum ‘with’ in Latin is equally wide: cum Pansa vixi in Pompeiano (‘I lived with Pansa at Pompeii’, Cic. Att. XIV.20.4); cum febri domum rediit (‘he came back home with a temperature’ Cic. Or. 3.6:); Habendum et ferendum hoc onus est cum labore (‘with pain this burden must be borne and endured’, Plaut. Am. 1. 1. 21); Ipsa satis tellus cum dente recluditur unco / Sufficit humorem, et gravidas cum vomere fruges (‘the earth itself, opened up by the curved ploughshare, gives enough moisture and heavy fruit’, Verg. G. II.423:); Qui cum Amphitruone abiit hinc in exercitum (‘and he (Sosia) with Amphitryon comes hither this day from the army’, Plaut. Am. prol. 125).



Exploring a diachronic (re)cycle of roles 141

when encoding some arguments, which in turn implies that the coding of one argument can shift from one regime to the other in both directions. Crucially, the area where punctual and relational coding overlap is provided by Prepositional Phrases, which are open to both coding regimes. In order to illustrate this point, compare, for instance, Prepositional Phrases that express grammatical relations under a relational coding regime and Prepositional Phrases that directly express conceptual relations under a punctual coding regime. An example of the former kind of structure is the prepositional object of intransitive two-place verbs: for instance, new cars must comply with high standards; I’m waiting for Joan. Examples of the latter are Location and Goal, which are arguments in the presence of stative predicates and verbs of motion respectively: Elizabeth lives opposite the Town Hall; I’m driving to Paris; John threw the ball beyond the fence. The difference of coding regime is immediately mirrored by the behaviour of the preposition. When it encodes a prepositional object, the preposition is selected by the verb and loses its semantic relevance (Steinitz 1969). Accordingly, it encodes a void grammatical relation, able to receive any content of role depending on the lexical semantics conveyed by the verb that governs it. By contrast, when it encodes a spatial relation, a preposition is chosen from amongst a large set, and directly contributes to the content of the role with its own transparent semantics. To clarify this point, let us compare the following expressions: (2) a. b.

Hans wartet auf einen Freund Hans wait.PRS.3SG on ART.ACC.M.SG friend ‘Hans is waiting for a friend’ Auf jedem Hügel liegt ein Turm on each.DAT.M.SG hill lie.PRS.3SG ART.NOM.M.SG tower ‘There is a tower on every hill’

In sentence (2a), the preposition does not contribute to the content of the role, which is directly determined by the main verb: there is no spatial relation, even metaphorical, between the two participants. The way it is assigned a role shows that prepositional object is a purely grammatical relation8 and an instance of relational coding. In sentence (2b), the preposition has a full lexical content, which directly takes part in defining the relevant spatial relation. This is the reason why the preposition is not 8.  A widely held idea is that prepositional coding is a clear indication of merely peripheral roles (Palmer 1994: 10). This is the view of Lazard (1998: 18), who claims that among arguments [actants] true and proper, those which take an oblique case (that is, not the zero or Accusative cases), or are marked by an adposition, are relatively peripheral; similarly, Kittilä et al. (2011: 9) maintain that core semantic roles are typically encoded by cases, while peripheral roles usually receive adpositional coding; and that the further one strays from the core roles, the more probable adpositional coding becomes.

142 Chiara Fedriani and Michele Prandi

selected by the verb but chosen by the speaker within a paradigm: Elizabeth lives in Warsaw (near the castle, opposite the University, and so on). The way the prepositional phrase receives a content shows that it is an instance of punctual coding. The fact that argumental Prepositional Phrases can operate under both relational and punctual coding regimes has some predictive power. As we shall illustrate in the following sections, the prepositional construction featuring ad and the Accusative, due later to take over the role of Latin Dative in Romance languages, starts its career as a typical form of expression of Goal encoded in a punctual way and ends up becoming the form of expression of the grammatical relation of indirect object under a relational coding regime that passes through a sequence of stages whose order can be exactly predicted and empirically documented. 3. The Dative role complex in Latin To begin with, the Latin Dative expresses the grammatical relation of indirect object, which is encoded in a relational regime. This is shown by the heterogeneity of roles encoded by it, which include the third argument of semantically different types of verbs as, e.g., do ‘I give’, trado ‘I transfer’, dico ‘I say’, scribo ‘I write’, and nego ‘I deny’.9 With verbs of delivering, the Dative encodes the Recipient: (3) totum denique hominem tibi ita trado entire.ACC.SG at.last man.ACC.SG 2SG.DAT so transfer.PRS.1SG ‘At last I transfer this man entirely to you’ (Cic. Fam. 7.5)

By contrast, with verbs such as dico ‘I say’, refero ‘I report’, ostendo ‘I show’, nuntio ‘I declare’, the Dative-marked argument conveys the role of Addressee. While Recipient and Addressee share some family resemblances, the third argument of such a verb as nego ‘I deny’ is completely eccentric: (4) Liviae (…) roganti civitatem negavit Livia.DAT beg.PART.PRS.DAT.SG freedom.of.the.city.ACC.SG deny.PF.3SG ‘He denied the freedom of the city to Livia, who was begging for it’ (Svet. Aug. 40)

In the presence of two-place verbs, the Dative typically expresses animate participants that are only marginally involved in the event, thus conveying the notion of indirect affectedness (Dąbrowska 1997; see also Næss 2007) and is also compatible

9.  Of course, Latin had more three-place verbs than these. For the purposes of the present study, however, it suffices to mention them.



Exploring a diachronic (re)cycle of roles 143

with different roles. In (5) the Dative encodes the Experiencer, while the Stimulus that causes the feeling is expressed as the subject: (5)

etenim mihi caligae eius et CONJ 1SG.DAT half-boots.NOM.PL POSS.3SG.GEN.SG CONJ fasciae cretatae non placebant band.NOM.PL marked.with.chalk.NOM.PL not like.IMPF.3PL ‘I don’t like his half-boots and his bands marked with chalk’ (Cic. Att. II.3.1)

Likewise, the Dative conveys the second argument of a number of verbs that tend to select non-transitive patterns of object realization cross-linguistically. They mainly express social interactions and psychological states where the Dative marks a human endpoint-like role, such as ‘to obey someone’, ‘to help someone’, ‘to threaten someone’, ‘to thank someone’ (see, e.g., Blume 1998). An example of such use of the Dative in Latin is provided in (6): (6)

ego autem ne irasci possum quidem 1SG.NOM CONJ CONJ get.angry.INF.PRS.dep can.PRS.1SG CONJ iis quos valde amo 3PL.DAT REL.ACC.PL greatly love.PRS.1SG ‘And for sure I cannot even get angry with those who I greatly love’ (Cic. Att. II.19)

Yet another use of this case within a two-place valency frame is to express possession within a construction featuring the verb esse ‘to be’, where the Dative-marked possessor is ascribed a possessum encoded as nominative subject: for instance, liber mihi est (‘I have a book’, lit. ‘a book medat is’). This is quite an ancient structure that displays several parallel counterparts in many other Indo-European languages (see, e.g., Ernout & Thomas 1953: 73, Baldi & Nuti 2010). In all of the abovementioned cases, the Dative is used to encode arguments which fit the valency frame of the main verb. However, the case also expressed a family of peripheral roles. When expressing peripheral roles, under the jurisdiction of a punctual coding regime, the Dative displays its typical content — namely, it encodes non-agentive and barely affected animate referents that are loosely involved in non-transitive types of relations, and thus simultaneously oriented toward animacy and non-agentivity (cf., e.g., Haudry 1968, Serbat 1996: Ch. 7, Van Lagendonck 1998, Menge 2007: 401–419). In the following examples the Dativemarked participants are not required by the valency frame of the relevant predicates, as they are margins that refer to interested parties.

144 Chiara Fedriani and Michele Prandi

(7) quid mihi Celsus agit? INT.ACC 1SG.DAT Celsus.NOM do.PRS.3SG ‘What is Celsus doing that is my concern? (I am particularly interested in what it is)’ (Hor. Ep. 1.3.15) (8) mihi discipulus … periit 1SG.DAT pupil.NOM.SG be.ruined.PF.3SG ‘My pupil is ruined/The pupil is ruined and this concerns me’ (Plaut. Bacch. 484)

These extra-argumental ‘free’ Datives signal involvement, special interest, and emotional participation and are clearly at the crossroads between the so-called Dative of interest, advantage, or concern (dativus commodi, dativus iudicantis) and the Dative of affect (dativus ethicus; cf. Hofmann 1936: 136–7, Ernout & Thomas 1953: 72, Woodcock 1959: 47, Hofmann & Szantyr 1972 [1965] II: 93–98, Serbat 1998: 564ff.). Given the affective value they convey, these adjuncts are often referred to in the literature as ‘empathetic’, ‘contact’, ‘interactional’, ‘affect’, ‘energetic’, ‘dynamic’ Datives (Van Hoecke 1996: 17–18, Menge 2007: 412–413, Fried 2010). The crucial point to be made here is that in such cases the Dative encodes contextbound, pragmatically-motivated margins that do not depend on the Noun Phrase or on the verb, but rather on the event as a whole (Van Langendonck 1996: 240), thus framing the content of a given state of affairs by emphasizing the speaker’s or the hearer’s point of view. In (7) and (8), for instance, the Dative case expresses the fact that the speaker (mihi) is especially concerned about the event described (discipulus periit, quid Celsus agit). So far, the functions typically taken over by Dative participants have been presented as clearly delineated and differentiated from one another. This picture by no means exhausts the possible discussion of Dative roles in Latin but it adequately serves to demonstrate that, on a syntactic level, they can fit two-place and threeplace valency frames as well as being added to the relational core with the function of margins, thus participating in both relational and punctual ways of coding. The arguments of both two- and three-place verbs — that is, Dative object and indirect object — are encoded in a relational way. Margins are encoded in a punctual way and share a common conceptual core: the expression of interested parties who are not directly affected by the event. In any case, there is no trace of the idea of motion towards a Goal in conceptual motivation, a point that will prove critical in our discussion. Before concluding, it is worth noting that role accumulation is also a common feature in the domain of the Latin Dative. There are basically two different kinds of role accumulation. First, one argument, whose semantic role is specified by the main verb, can be enriched by a sort of parasitic secondary role, which is either an

Exploring a diachronic (re)cycle of roles 145

Stimulus

Possessor

Goal

Recipient

Beneficiary/Maleficiary

Purpose

Addressee

Experiencer

Involvement

Figure 1.  A semantic map of the Latin Dative

implication of the main role or the outcome of an inferential process justified by some contextual elements or both. An example of such a type of accumulation of an argument and a peripheral role is (9) below, where we have the association between Recipient and External Possessor. It is self-evident that the Recipient is also the Possessor of the body part, which then receives extra characterization thanks to the reference to the subject’s hand. (9) Hanc mi in manum dat this.ACC.SG 1SG.DAT in hand.ACC.SG give.PRS.3SG ‘She placed her in my hands’ (Ter. Andr. 297)

Yet another typical case of accumulation involves Experiencer and External Possessor. Experiencers have been argued to be functionally connected to both the dativus ethicus, as they all imply some kind of affective involvement, surprise, or scorn, and to the dativus iudicantis, because Experiencers are also conscious witnesses (Serbat 1996: 572, Van Lagendonck 1998: 239, Luraghi 2010). In the following example, the Dative-marked participant mihi ‘to me’ assumes all of these functions, simultaneously carrying the role of Possessor (cf., the English translation ‘my’) and of Experiencer, as he is emotionally involved in the situation: as a matter of fact, the English translation does not do justice to the Latin Dative mihi ‘to me’, whose scope is widened to cover the effect of the verb also (‘my heart is tortured with my great concern’): (10) Cruciatur cor mihi et metuo torment.PRS.PASS.3SG heart.NOM.SG 1SG.DAT and fear.PRS.1SG ‘My heart is tortured and I fear’ (Plaut. Trin. 1169)

Example (10) above shows that the Possessor role points to semantically finer distinctions, as the participant is also especially concerned about the event (dativus ethicus) and is directly involved in the outcome of a negative situation, i.e., is damaged by having his heart broken, like a Maleficiary (dativus incommodi). In (9) and (10) the accumulation is asymmetric and hierarchical. However, different margins can add to each other, thus associating roles of the same rank. As examples, one can quote the accumulation of External Possessor and Beneficiary

146 Chiara Fedriani and Michele Prandi

or Maleficiary. The fact that the Possessor of an item can be seen either as a Beneficiary or a Maleficiary depending on the positive vs. negative characterization of the event (cf. Kittilä 2006) is clearly acknowledged in the following examples, which are structurally equivalent: the subject is constituted by a body part in relation to the External Possessor mihi ‘to me’. In (11), the relation is clearly positive (eyes shine) and as a result the Possessor is also concerned, in a beneficial state. In (12), by contrast, the participant bears the semantic determinant of malefaction, as he is negatively affected by his teeth falling out. (11)

pleni oculi sorderum qui erant, full.NOM.PL eye.NOM.PL dirt.GEN.PL REL.NOM.PL be.IMPF.3PL iam splendent mihi now shine.PRS.3PL 1SG.DAT ‘My eyes which were full of dirt, they shine (for me) now’ (Plaut. Poen. 314)

(12) tunc dentes mihi cadebant primulum then tooth.NOM.PL 1SG.DAT fall.out.IMPF.3PL at.first ‘my teeth were falling out (of me) for the first time (with my concern)’ (Plaut. Men. 1116)

It should be stressed that the Possessor is not an argument, but a relation involving one argument: in (12), for instance, the process expressed by dentes cadebant ‘my teeth were falling out’ is saturated by all of the arguments required by its valency frame. The expression of the External Possessor does not change the conceptual structure of the process but enhances the salience of the participant involved. Summing up, the Dative case in Latin performed a wide range of functions, both in the domain of relational coding (the grammatical relation of indirect object) and in the realm of punctual coding (margins, optional free Datives). Moreover, it was used to express the second argument of a variety of verbs indicating human relationships (ex. 6–7), in which, although marking a relationally established argument, the choice of the case was at least partially semantically motivated, being limited to animate and non-agentive participants. This network of functions is illustrated in the semantic map in Figure 1, which shows how the Dative role complex in Latin was composed. Building on this characterization of the Dative role complex in Latin, in the next section we go on to focus on the process of diachronic replacement that led the prepositional strategy with ad ‘to’ to progressively intrude into the functional domain of the Dative case, thus paving the way for it to finally get the upper hand in the Romance languages.



Exploring a diachronic (re)cycle of roles 147

4. The diachronic development of the Dative role complex in Latin In this section we will consider the occurrence of ad ‘to’ to express Dative-like roles having both the status of arguments and margins within a chronological span ranging from Plautus to the Vulgate (5th century CE). The basic aim of this section is to explore the paths along which the prepositional strategy gradually got a foothold in taking over the functions traditionally associated with the Dative role complex. More recently, various linguists have looked at this alternation and suggested some semantic distinctions; however, although valuable observations have been made, they usually confine themselves either to verb-specific analyses or to single authors. For instance, Théoret (1982) and Baños Baños (1996, 1998) only take into account the alternation in Cicero and in Classical Latin, Baños Baños (2000) focuses on Plautus, Sznajder (2012) on verbs of saying in the Vulgate, while Pinkster (1990), Selig (1991), and Van Hoecke (1996), although providing important insights as to the semantic motivations that constrained the expansion of ad ‘to’ at the cost of the Dative case, do not provide a wide-ranging and text-based diachronic account. Our aim is therefore to provide such an analysis: more specifically, we try to establish i. up to what cut-off point Latin more forcefully maintains the Dative within the role complex, and when, by contrast, the competing prepositional strategy begins to prevail; and, ii. whether there are plausible generalizations to be made about preferred semantic paths of extension. To accomplish this goal, we will be verifying whether some classes of verbs are more likely to accommodate the alternative prepositional construction than others, and whether our results can be arranged in a semantic hierarchy. In doing so, we will gain some insight into how to account for the data by exploring the micro-dynamics through which this diachronic replacement took place, causing an ancient role complex expressed by a relational case to give rise to a new role complex in the Romance languages by passing through a phase of punctually motivated coding (Section 5). The investigation of this process of diachronic substitution sheds particular light on the competition between conflicting strategies underlying the organization of the grammatical system. In point of fact, the distance between the Dative and the prepositional strategy could not be wider: when encoding the indirect object, the Dative instantiates the ideal type of relational coding, whereas the competing prepositional strategy is a typical instance of semantically motivated punctual coding. Our procedure in this investigation will therefore be to analyze the occurrence of ad ‘to’ plus the Accusative with three-place verbs (do ‘I give’, mitto ‘I send’, dico ‘I say’, nego ‘I deny’), with two-place verbs selecting for the DAT-NOM case

148 Chiara Fedriani and Michele Prandi

frames (mihi videtur ‘(it) seems to me’), as well as with margins (e.g., constructions featuring an ethical Dative). Our diachronic corpus consists of the following works: Plautus’ comedies for Early Latin (3th century BCE), Cicero’s letters for Classical Latin (1th century BCE), and the Vulgate and the Itinerarium Egeriae (5th century CE) as evidence for Late Latin. Before discussing the data, however, some methodological remarks are in order. Importantly, the data reveal that the extension of the innovative form featuring ad ‘to’ plus the Accusative does not involve the whole Dative complex outlined in Section 3, but only a subset of three-place verbs, leaving aside two-place verbs and margins. In order to account for these data, one has to ask some relevant questions, namely: why are some arguments of some verbs under attack in Latin whereas some others are not? Why do some arguments surrender earlier than others? Why are some roles absolutely reluctant to take the innovative form while suddenly accommodating it in the very early stages of the development of the Romance languages? Why do margins forcefully maintain their Dative form in Latin, whereas in the Romance languages the innovative form is extended even beyond the boundary with the Latin Dative complex? Our claim is that the drift follows a path of decreasing conceptual motivation and increasing cognitive cost. More specifically, if we arrange the verbs involved in the change along an ideal cline, we first find those that imply a concrete motion across space. The most typical case is mitto ‘I send’. In this case, what is required in order to justify the innovative form is a recategorization of the Addressee of the message as a metaphorical Goal, a move which is in turn prompted by the metonymic connection between the Addressee and the place s/he occupies, i.e., the actual location constituting the Goal. Instead of being simply connected with the place identifying the Goal, the Addressee metaphorically becomes this very Goal: (13)

misine ego ad te ex Epheso send.PF.1SG-PTC 1SG.NOM to 2SG.ACC from Ephesus.ABL epistulam super amica (…) ? letter.ACC.SG about mistress.ABL.SG ‘Did I not send you a letter from Ephesus about my mistress?’ (Plaut. Bac. 561)

By contrast, in the presence of a verb such as scribo ‘I write’ used in its communicative sense (‘to write to somebody’), motion is not an inherent feature implied by the process, but is connected to it through a metonymic link. Thanks to a metonymic shortcut, such a complex chain of events as to write a message on a medium and send it to the place where someone lives is condensed into the expression to write a message to someone, which both transfers onto write the valency scheme of send and categorizes the Addressee as if it were a metaphorical Goal:



Exploring a diachronic (re)cycle of roles 149

(14) ibi sedens haec ad te scribebam here sit.PART.PRS.NOM.SG this.ACC.N.PL to 2SG.ACC write.IMPF.1SG ‘Sitting there as I was writing these things to you’ (Cic. Att. I.10)

The last stage in this development is constituted by dico ‘I say’, where motion is neither real nor metonymically connected to the event. In order to recategorize the Addressee as a metaphorical Goal, the process itself as a whole has to be reinterpreted metaphorically as a kind of motion: (15)

Ait autem ad illos Iesus: Interrogo say.PRS.3SG then to that.ACC.PL Jesus.NOM ask.PRS.1SG vos si licet sabbatis benefacere, an male? 2PL.ACC if be.lawful.PRS.3SG on.the.sabbath to.do.good.INF.PRS or evil ‘Then said Jesus to them: I will ask you one thing. Is it lawful on the Sabbath days to do good or to do evil?’ (Luke 6: 9–10)

Returning to the questions above, our sketch reveals significant predictive power. On the one hand, the diachronic drift is expected to follow a well-defined path: the verb dico ‘I say’, for instance, should be the last to adopt the innovative form, whereas mitto ‘I send’ and scribo ‘I write’ are expected to enter the innovative construction earlier than dico ‘I say’. On the other hand, no change is documented outside this constellation of processes in Latin. The reader may have noted that we have yet to mention the verb do ‘I give’, which is often regarded as highly prone to accommodating the prepositional strategy at an earlier stage. As we will show, do ‘I give’ fits our predictions, but its behaviour is more complex than one would imagine at first glance. For this reason, our detailed analysis of this verb sheds light on many theoretical and empirical issues which have important consequences for a better understanding of the entire diachronic cycle involving the Dative role complex. Let us now take as a starting point our Early Latin data. As mentioned briefly in Section 1, ad ‘to’ replacing the Dative is used as early as in Plautus. Interestingly, however, in Early Latin it is exclusively attested in the expression of arguments with verbs of giving (ex. 16), sending (ex. 17), or the dispatching of letters (ex. 18), which all entail the targeted motion of an object across space and presuppose a focus on reaching the final destination (see Adams 2011: 266 for further exemplification). (16)

Si in singulis stipendiis is if in each.ABL.PL campaign.ABL.PL 3SG.NOM ad hostis exuvias dabit to enemy.ACC.PL spoils.ACC.PL give.FUT.3SG ‘If in each campaign he yields spoils to the enemy’ (Plaut. Epid. 37)

150 Chiara Fedriani and Michele Prandi

(17)

nam ego hodie ad Seleucum regem misi for 1SG.NOM today to Seleucus.ACC king.ACC.SG send.PF.1SG parasitum meum parasite.ACC.SG POSS.1SG.ACC.sg ‘For this day I have sent my parasite to King Seleucus’ (Plaut. Mil. 947)

(18)

misine ego ad te ex Epheso send.PF.1SG-PTC 1SG.NOM to 2SG.ACC from Ephesus.ABL epistulam super amica (…)? letter.ACC.SG about mistress.ABL.SG ‘Did I not send you a letter from Ephesus about my mistress?’ (Plaut. Bac. 561)

The fact that the extension of ad and the Accusative is limited in Early Latin to expressing concrete transfers of objects from a Source to a Goal is due to the fact that the human Addressee and Recipient are metaphorically construed as the final Goal of the transaction. Therefore, the Direction schema is mapped onto such valency frames and the punctual coding featuring ad ‘to’ readily presents itself. More specifically, our data show that the prepositional strategy was gradually extended where it was perceived as semantically transparent, that is, as evoking the idea of concrete transport. What we essentially have here is a combination of metonymic shifts and metaphorical transfers whereby the role of Goal (typically a location) is projected onto the Addressee and Recipient under a punctual coding regime. Along this line of reasoning, Baños Baños (1996: 227ff.) has neatly shown that the prepositional strategy is always preferred in Cicero when letters or persons (usually carrying messages) are transferred across space. By contrast, when the Recipient is given money, presents, or abstract entities like salus ‘salvation’ (Cic. Sul. 23), the transfer schema does not apply and the Dative is usually selected. We suggest that this is due to the fact that letters and messages logically imply a source for their existence, namely a writer who represents the Source and is placed in a location which, as a rule, differs from that occupied by the Recipient. This is why the Recipient can easily be mapped as the final target, thus facilitating the compatibility with the motion schema implied by ad ‘to’. By contrast, other types of entities like money or gifts are typically delivered in praesentia and do not necessarily imply a concrete motion in order to be transferred from one place to another. Therefore, the targeted participant is more reluctant to be accommodated within the Goal-like valency frame and is most often assigned the Dative case. We therefore claim that there is a clear difference between in absentia and in praesentia transfers, which is of the utmost importance in explaining the diachronic cycle of the Dative complex, as these two categories greatly differ as to their implication of motion as a necessary semantic component. While in absentia transfers always imply a real movement across space (e.g., the event of dispatching



Exploring a diachronic (re)cycle of roles 151

a letter from Rome to Greece), this is not the case in typical in praesentia transfers, where something is delivered from one’s hands to the Recipient’s without actual displacement (e.g., in the case of personally giving a present to a friend who stands in front of one). As we shall see, this semantic distinction is grammatically relevant in that it triggers and constrains the use of alternative coding options and gives rise to a diverging distribution, which is not random but corresponds to a ratio that matches our predictions. Let us briefly expand on this important point by again looking at the case of letters, which generally implies a transfer across space and easily accommodates the prepositional schema. We first focus on the behaviour of the verb do ‘I give’, which is very telling and helps to shed light on the issue. As early as in Plautus the prepositional strategy is documented with some uses of this verb (cf. ex. 1b) while the Dative form is preserved in other contexts (cf. ex. 1a). The data reveal that in particular cases where the motion schema is not activated, the Recipient takes the Dative case. In the following examples, an Agent personally hands a letter — that is, an object — to a Recipient, who is entrusted with the personal delivery of the letter to the final Addressee of the message. Interestingly, the Dative is selected to encode the in praesentia Recipient of the object (cui ‘to whom’ in ex. 19, and Philotimo ‘to Philotimus’ in ex. 20, respectively) while the prepositional strategy is used to express the final in absentia Addressee of the message, which implies a delivery from a Source to a spatially distant endpoint (ad te ‘to you’ in both examples). (19)

cum primum Romam veni fuitque cui at.first Rome.ACC arrive.PF.1SG be.PF.3SG-CONJ REL.DAT.SG recte ad te litteras darem safely to 2SG.ACC letter.ACC.PL give.SUBJ.IMPF.1SG ‘As soon as I reached Rome and there was anyone to whom I could safely entrust a letter for you’ (Cic. Att. IV.1)

(20)

exiens e Pompeiano Philotimo go.away.PART.PRS.NOM.SG from Pompey.ABL Philotimus.DAT dederam ad te litteras give.PPF.1SG to 2SG.ACC letter.ACC.PL ‘As I was leaving my Pompeian villa, I had delivered a letter for you to Philotimus’ (Cic. Att. V.3)

The data just discussed support Baños Baños (1998) and Luraghi’s (2010) claim that ad ‘to’ with the Accusative is primarily used when Recipients are metaphorically mapped onto Goals and that the semantic entailment of motion is a necessary prerequisite that paved the way for the extension of this strategy to new contexts. Another example that clarifies this opposition is (21), where the Dative is chosen to encode an in praesentia Recipient (L. Caesari ‘to L. Caesar’), who has

152 Chiara Fedriani and Michele Prandi

to report a message to an in absentia Addressee, who, as one may have expected, is expressed by ad ‘to’ with the Accusative (ad illum ‘to him’): (21)

nunc has exspectationes habemus now this.ACC.F.PL expectation.ACC.F.PL have.PRS.1PL duas, unam quid Caesar two.ACC.F one.ACC.F INT.N Caesar.NOM acturus sit cum acceperit ea do.ACT.PERIPHR.NOM.SG.M when receive.FUT2.3SG this.ACC.N.PL quae referenda ad illum REL.ACC.N.PL report.grv.ACC.N.PL to 3SG.ACC data sunt L. Caesari give.PF.3PL.PASS L.Caesar.DAT ‘Now we are in suspense on two points: first, what Caesar will do when he has received the answer entrusted to L. Caesar to report to him’ (Cic. Att. VII.16)

Example (22) provides further evidence for this semantically constrained alternation. In this revealing passage, the very same verb of delivering, once again do ‘I give’, is repeated twice, but realizes two different argument marking patterns depending on the presence or lack of entailed motion. A letter is first entrusted to an in praesentia Recipient, which takes the Dative case; the in absentia Addressee is subsequently encoded instead through the prepositional strategy (ad te ‘to you’). (22)

dederam equidem L. Saufeio litteras et give.PPF.1SG in.fact L. Saufeius.DAT letter.ACC.PL and dederam ad te unum give.PPF.1SG to you.ACC alone.ACC.M ‘I did, in fact, give L. Saufeius a letter, and to you alone’ (Cic. Att. VII.1)

Further evidence for the metaphorical projection of the concept of destination onto the ad-marked Recipient comes from a detailed analysis we have carried out on the specific contexts associated with mitto ‘I send’, do ‘I give’, and scribo ‘I write’ with litteras and epistulam/epistulas ‘letter(s)’ in Cicero’s letters to Atticus. Interestingly, our data show that the prepositional strategy is more frequent with the indicative mood and/or in perfective contexts, where a special focus is cast on successfully reaching the final Goal. Although the perfective and realis features are by no means exclusively associated with the prepositional strategy, it is revealing that in irrealis or imperfective contexts, i.e., when reaching the destination is only suggested, imagined, or explicitly negated, almost only the Dative-marked counterpart is admitted (21 occurrences vs. three). In these cases, the Recipient cannot indeed be easily conceptualized as a metaphorical reachable destination. Our findings are quantitatively summarized in Table 1.

Exploring a diachronic (re)cycle of roles 153



Table 1.  Relative frequency of occurrence of Dative vs. ad + Acc. depending on (ir)realis mood and (im)perfective aspect in Cicero’s letters Ad + Acc.

Dative

Indicative mood and/or perfective aspect

57

22

Subjunctive mood and/or irrealis reading

 3

21

Total

60

43

Remarkably, the very same tendency emerges if the analysis is extended to the whole predicate, including a typology of entities that are delivered. To illustrate this crucial point further we have compared the constructional behaviour of nuntius ‘message’ and munus ‘gift’ as direct objects, which illustrate prototypical in absentia and in praesentia recipiental relations, respectively, in the Vulgate, to obtain immediate evidence of the diachronic evolution from the situation presented above for Cicero onwards. The data drawn from the Vulgate clearly testify to a persistent difference in coding: while the in absentia Recipient of messages is always expressed by means of ad and the Accusative in this work, the in praesentia Recipient of gifts is rather encoded using the Dative. Our findings are summarized in Table 2. Table 2.  Complementary distribution in the coding of in praesentia vs. in absentia Recipients in the Vulgate Dative

Ad + Accusative

nuntius ‘message’

 –

34

munus ‘gift’

35

 2

Very interestingly, in the only two cases in which the Recipient of munus is unexpectedly expressed using the prepositional strategy in the Vulgate, it is coordinated either with litteras ‘letters’ (ex. 23) or with libros ‘books’ (Isaiah 39), which were sent across space and dispatched to prototypical in absentia Recipients. This is why we find the constructional pattern overwhelmingly associated with recipient human Goals: (23)

in tempore illo misit Berodach in time.ABL.SG that.ABL.SG send.PF.3SG Berodach.NOM Baladan filius Baladan rex Baladan.NOM son.NOM.SG Baladan.GEN king.NOM.SG Babyloniorum litteras et munera ad Ezechiam Babylon.GEN letter.ACC.PL and gift.ACC.PL to Ezekiah.ACC ‘At that time Merodach-Baladan son of Baladan king of Babylon sent Hezekiah letters and gifts’ (Kings 2, 20:12)

154 Chiara Fedriani and Michele Prandi

In sum, we have provided diachronic evidence to support the claim that in praesentia vs. in absentia transfers do motivate different forms of coding until Late Latin and should be kept apart. The historical process which resulted in the systematic substitution of the Dative case by ad ‘to’ plus the Accusative followed different paths that triggered the acquisition of the new pattern of coding at different periods. The points just discussed will guarantee a better understanding of the kinds of verbs subsequently admitted to feature the competing prepositional strategy. After verbs of giving and sending, in Classical Latin we find verbs of writing (to someone). Note that they are placed at a higher level of abstraction, in that in such a case a process of motion is only indirectly suggested. As we have remarked above, however, we are dealing in this case with an interesting synergy between a metaphoric transfer and a metonymic shift. What makes the metaphorical categorization of the Addressee as a Goal possible is a logically prior metonymic shift which cuts across a complex chain of events: to write a message on a medium and send it to the place where someone lives is condensed into to write a message to someone. The outcome of this process appears in Cicero: note that the prepositional turn is employed both when the physical medium is explicitly mentioned (litteras ‘letters’: ex. 24) and when it is not overtly expressed and the verb directly governs the prepositional phrase (ex. 25): (24)

a te litteras crebro ad me from 2SG.ABL letter.ACC.PL frequently to 1SG.ACC scribi video write.INF.PRS.PASS notice.PRS.1SG ‘I notice that you have frequently written letters to me’ (Cic. Att. IV.6)

(25) scripsi etiam ad Camillum, ad Lamiam write.PF.1SG also to Camillus.ACC to Lamia.ACC ‘I have written also to Camillus and Lamia’ (Cic. Att. V.8)

Crucially, verbs of written communication conform to the conceptual scenario depicted above, as they fulfill all of the semantic requirements characterizing verbs of giving and sending: by means of a metonymic shortcut, they entail a movement across space, and they presuppose a delivery in absentia, thus favouring the metaphorical mapping of the Goal to be attained onto the Addressee. Table 3 summarizes the constructional behaviour of do ‘I give’, mitto ‘I send’, and scribo ‘I write’ taking litteras and epistulam/epistulas ‘letter(s)’ as objects in Cicero’s letters to Atticus.

Exploring a diachronic (re)cycle of roles 155



Table 3.  Alternation in coding of Recipients depending on the kind of transfer in Cicero’s letters to Atticus Ad + Acc.

Dative

Typical kind of transfer

do ‘I give’

29

36

in praesentia

mitto ‘I send’

27

16

in absentia

scribo ‘I write’

11

 3

(metonymically) in absentia

Note that when the prepositional variant took hold, it rapidly encroached on the function previously reserved for the Dative, especially with verbs that presuppose a transfer in absentia, namely mitto ‘I send’ and scribo ‘I write’. With do ‘I give’, that is, the quintessential delivering verb, the motion from a source to a goal is not very prominent from a conceptual point of view, because the act of giving something to someone is typically performed personally, from the hands of the giver to those of the Recipient, without necessarily entailing a concrete displacement across space. The lack of metaphorical mapping of the Goal onto the Recipient therefore results in a similar distribution of case-marked and prepositionally marked Recipients with this verb (36 vs. 29), with no clear preference for the prepositional turn over the use of the Dative case, and a higher frequency of this latter strategy instead. A closer look at this distribution, however, clearly reveals that the selection of one strategy over the other is constrained by precisely the same semantic parameters, i.e. the presence vs. the lack of movement. More precisely, ad plus the Accusative is typically selected (27 times out of 29) to convey in absentia deliveries with a clear semantic coercion of the verbal meaning: in these cases, litteras/epistulas dare always means ‘to write, to dispatch, to entrust a letter’, with the prepositional phrase playing a crucial role in enhancing and determining a semantic shift from an in praesentia to an in absentia transfer.10 A typical example is given in (26), where do + ad and the Accusative means ‘I dispatch to’: 10.  Importantly in the present context, this systematic semantic shift is clearly testified by Loeb’s translations into English of the relevant contexts. When the prepositional strategy is combined with dare litteras/epistulam, epistulas, the English equivalent is ‘to send, dispatch, write a letter/ letters (to someone)’. Besides example (26), some more illustrative passages, among many others, are ad Octavium dedi litteras ‘I have written to Octavius’ (lit. ‘to Octavius I gave letters’: Att. II.1), Tiro ad te dedisset litteras ‘Tiro would have written to you’ (lit. ‘Tiro would have given a letter to you’: Att. VI.7), cum dedissem ad te litteras tristis ‘After I had sent to you a gloomy letter’ (lit. ‘after I had given to you a gloomy letter’: Att. VII.25), cum ad te litteras dedissem, redditae mihi sunt litteras a Pompeio ‘after I have dispatched a letter to you, I received one from Pompey’ (lit. ‘after I have given a letter to you, I received one from Pompey’: Att. VIII.1), litteras ad te de Dyonisio dedissem ‘having written you a letter about Dyonisius’ (lit. ‘having given you a letter about Dyonisius’: Att. VIII.5), si tibi videretur, ad me litteras dare ‘if you think it well to write to me’ (lit. ‘if you think it well to give letters to me’: Att. VIII.11b).

156 Chiara Fedriani and Michele Prandi

(26)

cum dedissem ad te litteras ut CONJ give.SUBJ.PPF.1SG to 2sg.ACC letter.ACC.PL CONJ scires Caesarem Capuae vii Kal. fore know.SUBJ.IMPF.2SG Caesar.ACC Capua.GEN 26th be.INF.FUT ‘After I had dispatched the letter to you informing you that Caesar would be at Capua on the 26th’ (Cic. Att. IX.15)

By contrast, when do ‘I give’ selects a Dative-marked Recipient it always carries its literal meaning pointing to in praesentia transfers. This clearly testifies to the lack of metaphorical mapping interpreting the Dative-marked participant in terms of a figurative Goal. The detailed frequencies emerged from a thorough analysis of do ‘I give’ + litteras and epistulam/epistulas ‘letter(s)’ in Cicero’s letters are summarized in Table 4, with bold boxes highlighting the preference of ad + Acc. for in absentia transfers (‘to write, to dispatch a letter to someone’) and the selection of the Dative case for in praesentia transfers. Table 4.  Alternation in coding of Recipients with do ‘I give’ in Cicero’s letters to Atticus Ad + Acc.: 29 occurrences do ‘I give’

Dative: 36 occurrences

In praesentia

In absentia

In praesentia

In absentia

2

27

35

1

One interesting matter which has not yet been raised, however, is to determine along which paths the prepositional variant was gradually extended to new verbs, thus also admitting abstract transfers in praesentia. The data from our corpus reveal that this further step is the extension of the innovative valency scheme from verbs of writing to verbs of oral communication. In this case, it took longer for the prepositional strategy to gain the upper hand over the Dative form: the complete conflation of the Addressee of a verb of speaking and the typical coding frame of Goal is only attested at a later stage, by the 5th century CE, in the Itinerarium Egeriae (ex. 27) and in the Vulgate (ex. 28), where the Greek model eipeîn prós tina ‘to say to someone’ probably played a key role in reinforcing the selection of the prepositional scheme (for an exhaustive discussion of this issue, see Sznajder 2012). (27)

Sanctus episcopus, ait ad me: holy.NOM.SG bishop.NOM.SG say.PRS.3SG to 1SG.ACC Eamus nunc ad portam go.SUBJ.PRS.1PL now to door.ACC.SG ‘The holy bishop said to me: let us now go to the door’ (Per. XIX.16)



(28)

Exploring a diachronic (re)cycle of roles 157

dicit mater Iesu ad eum say.PRS.3SG mother.NOM.SG Jesus.GEN to 3SG.ACC vinum non habent wine.ACC.SG not have.PRS.3PL ‘The mother of Jesus says to him they have no wine’ (John 2: 3)

This extension of the motion valency frame to verbs of saying has been convincingly explained by Luraghi (2010) in the light of the “conduit” metaphorical schema (Reddy 1993; see also Lakoff 1987: 450–451), whereby words are conceived of as objects moving along a path from a Source (the Speaker) to a Goal (the Listener). As a result, we have a process of constructional generalization in view of which the prepositional turn with ad ‘to’ is reinterpreted as part of the prototypical ditransitive valency scheme featuring a generic Goal of messages and words, which can be delivered either by means of a concrete support (cf. the case of mitto ‘I send’) or “in the air” (as with dico ‘I say’; note that scribo ‘I write’ has played an important role in this drift thanks to the metonymic shortcut, as documented above). Owing to the same kind of metaphorical transfer, the event of saying something to someone amounts to the valency frame of giving in many languages (see e.g. Newman 1996: 89, 138 for an overview; Kittilä 2006 on Finnish, Jacob 1991 on French, Nørgård-Sørensen, Heltoft & Schøsler 2011: 250–258 on Old French and Old Occitan). A further step in our research has been to check whether the prepositional strategy had been extended in Late Latin to those verbs that are not compatible, even metaphorically or metonymically, with the scheme of motion towards a Goal. We have thus examined the constructional options allowed by the verbs nego ‘I deny, refuse (something to someone)’ and videtur ‘it seems (to someone)’, where a void grammatical relation receives content from the relational meaning of the main verb that does not have an allative component as part of its semantics. Our data reveal that the use of ad ‘to’ has not replaced the Dative in these cases, which remain firmly anchored in a relational form of coding (but see Molinelli 1998 for some interesting examples and discussion). The indirect object of nego ‘I deny’, for instance, is always encoded by the Dative case even as late as in the Vulgate: (29) neque enim negare tibi quicquam potest not for deny.INF.PRS 2SG.DAT it.ACC can.PRS.3SG ‘for he will not deny it to you’ (Kings 1, 2)

To conclude, the arguments we discussed in the preceding pages seem to account for the data in a satisfactory way. In Latin, the extension of ad ‘to’ is constrained by a step-by-step dynamic process along the lines of metaphorical transfers and metonymic shifts which are always punctually motivated, whereas the verbs inaccessible to this conceptual motivation remain firmly anchored in a relational coding

158 Chiara Fedriani and Michele Prandi

regime. Our findings are summarized in Table 5 below, which illustrates the occurrence of ad ‘to’ and the Accusative to express the Recipient/Addressee/Listener mapped onto a Goal with the verbs analyzed at different diachronic stages. Table 5.  Use of ad and the Accusative with different classes of verbs do ‘I give’

mitto ‘I send’

scribo ‘I write’

dico ‘I say’

nego ‘I deny’

videtur ‘it seems’

Plautus Cicero Egeria Vulgate

The last step in our diachronic survey was to investigate whether the prepositional strategy is attested with margins, for example when replacing so-called ethical Datives. According to our predictions, since such contexts do not imply motion, not even in metonymic or metaphorical senses, we always find Datives expressing such optional adjuncts even in Late Latin (pleonastic reflexives in the terminology of Cennamo 1999, who cites interesting cases from many late texts, including the one given as 30): (30)

Ite interim nunc unusquisque ad domumcellas go.IMP.PRS.2PL meanwhile now each.NOM.SG to house.ACC.PL vestras, sedete vobis et modico POSS.2PL.ACC.PL sit.IMP.PRS.2PL 2PL.DAT and awhile ‘Go now, each one of you, to your houses, and sit down awhile’ (Per. XXXVI.5)

(31)

et transite vobis habitatio pulchra, and go.IMP.PRS.2PL 2PL.DAT home.NOM.SG beautiful.NOM.SG confusa ignominia bewild.PART.PF.NOM.SG shame.ABL.SG ‘And go, your beautiful home is bewildered with shame’ (Vulg. Mich. I. 11)

Summing up, the main point that emerges from our discussion is that the gradual replacement of the Dative with the prepositional strategy is diachronically motivated by a metaphorical mapping featuring the Recipient and Addressee as if they were metaphorical Goals. This testifies to the fact that the competing prepositional schema is semantically transparent since its use is restricted to verbs implying some kind of transfer as an integral part of their semantics and does not apply to other verbs, such as deny and seem, or to margins. This presupposes a form of coding built on a clear conceptual motivation, where the preposition still retains its



Exploring a diachronic (re)cycle of roles 159

spatial meaning, thus imposing constraints on those contexts of occurrence that show full semantic compatibility with it. In the next section we take into account the final stage that concludes the (re) cycle of the Dative role complex from Latin to Romance and discuss later developments whereby ad ‘to’ came to fully replace the Dative case in the complete range of its former functions. The price to be paid for this turn, as we shall see, is a loss in transparent semantic motivation and a shift of the prepositional strategy towards a relational coding regime. 5. Completing the (re)cycling of the role complex: Evidence from the Romance languages The balance resulting from the struggle between the Dative case and its innovative competing construction in Late Latin presupposes the full vitality of the case system. Therefore, it was destined to collapse with the disappearance of case morphology in Early Romance languages (except Romanian, where a two-case distinction, namely nominative-accusative vs. genitive-dative, is still preserved), when the ‘Dative’ function was fully replaced by the prepositional strategy, being only occasionally expressed synthetically for pronouns. This, it should be stressed, is not a trivial datum. In English, for instance, in the absence of any differential case marking, the marking of the indirect object by means of a preposition competes with its marking by means of its position before the direct object: Mary gave a pen to Kitty; Mary gave Kitty a pen. At any rate, the relevant point is that the prepositional strategy is not limited to the uses documented in Latin but spreads to verbs which do not imply any real or figurative motion from a Source to a Goal, thus becoming open to roles other than the metaphorical goal of a transaction. This provides evidence for the claim that this construction gradually came to be entitled to express the grammatical relation of indirect object, which, like any other grammatical relation, receives its content from the main verb and is not bound to a restricted family of roles that are semantically compatible with the original spatial meaning conveyed by the preposition. As early as Old Italian and Old French, the indirect object coded by means of a/à + NP is no longer limited to verbs of giving and communication. Though sharing the same ditransitive valency scheme, such verbs as Old Italian sottomettere ‘to subdue’ (ex. 32), paragonare ‘to compare’ (ex. 33), Middle Old French préférer ‘to prefer’ (ex. 34), and Old Spanish anteponer ‘to prefer’ (ex. 35) shape processes which cannot be reduced, even metaphorically, to the model of an allative transaction, and therefore assign to the indirect object a role that has nothing allative about it.

160 Chiara Fedriani and Michele Prandi

(32)

E così cominciò il Comune di Firenze and so begin.PAST.3SG the city of Florence a distendersi (…) sottomettendosi a la giuridizione to grow.INF.PRS subdue.GER.PRS to the jurisdiction ogni nobile di contado every noble of district ‘And thus the city of Florence began to grow (…) and they subdued to their jurisdiction every noble of the district’ (Giovanni Villani, Nuova Cronica 4.36; 14th century)

(33)

non v’ha sotto il sole terra che possa not there is under ART.M.SG sun land that can.SUBJ.PRS.3SG paragonarsi all’ Italia compare.INF.PRS-RFL to Italy ‘There’s no land under the sun that can be compared to Italy’ (Petrarca, Epistole senili VII; 14th century)

(34)

Il est vray; mais un bien particullier it be.PRS.3SG true but a good single n’est pas à preferer à la chose publique NEG to prefer.INF.PRS to the thing.SG public.SG ‘It is true; but a private good is not to be preferred to public interest’ (Jean de Bueil, Le Jouvencel, t.2, 12; 15th century)

(35)

En que nos da a entender que las in which 1PL give.PRS.3SG to understand that ART.F.PL çelestiales son de anteponer a las terrenales celestial be.PRS.3PL to prefer to earthly ‘In which it makes us understand that celestial issues must be given preference to the earthly ones’ (Anonymous, Un sermonario castellano medieval, Sermón VII; 15th century)

These examples, dating back to the ancient layers of the languages involved, clearly show that with the collapse of case systems for nouns, a/à + NP lost its semantic import and was successfully recruited to express the entire constellation of functions previously conveyed by the Latin Dative. Other verbs meaning ‘to (dis)please’ joined the same construction in Old Italian and Old French: agreer, atalenter, (des)plaisir, ennoyer, loisir, nuisir, seoir; ‘to deplore, regret’: peser, grever, doloir (see Mathieu 2006 for a complete list), along with many other verbs of experience that do not imply any kind of transfer, such as Old French and Old Italian covenir/convenire ‘to suit/to be convenient’, chaloir/calere ‘to matter’, avenir/avvenire ‘to happen’, and others.



Exploring a diachronic (re)cycle of roles 161

The Dative complex documented in the modern Romance languages simply confirms this early stage. The range of verbs that take the form a/à + NP to encode the indirect object is far wider and more heterogeneous than in Latin. Besides verbs of giving and communication, modern Romance languages entrust to the indirect object the expression of the third argument of large families of verbs whose relational content has nothing to do with motion even in a metaphorical sense. Among such verbs, we have It. invidiare, Sp. envidiar, Fr. envier ‘to envy’, It. impedire and negare, Sp. empedir, Fr. nier ‘to prevent, deny’, It. paragonare, Sp. comparar, Fr. comparer ‘to compare’, It. permettere, Sp. permitir ‘to allow’, It. anteporre and preferire, Sp. anteponer and preferir, Fr. antéposer and préférer ‘to prefer’, among many others. It is worth stressing in this context that some of the verbs mentioned do not even impose the condition of animacy on the referent of the indirect object. This holds, for instance, for It. paragonare, Sp. comparar, Fr. comparer ‘to compare’ and It. anteporre and preferire, Sp. anteponer and preferir, Fr. antéposer and préférer ‘to prefer’. It is readily apparent from the examples given above that the category of indirect object in Romance languages is not bound to a consistent family of roles, but clearly displays the status of a void grammatical relation that can be semantically filled in different ways according to the relational content of the verb that governs it. A revealing case in this respect comes from the behaviour of the French converse verbs prêter ‘to lend’ and emprunter ‘to borrow’. Both verbs share exactly the same valency scheme regardless of the fact that they impose two opposite perspectives on the same transaction, thus conferring two opposite roles on the indirect object, that is, the roles of Goal and Source: (36) a. b.

Jean a prêté mille francs à Luc Jean lend.PAST.3SG one.thousand francs to Luc ‘Jean lent one thousand francs to Luc’ Luc a emprunté mille francs à Jean Luc borrow.PAST.3SG one.thousand francs from Jean ‘Luc borrowed one thousand francs from Jean’

The presence of such a heterogeneous constellation of verbs accommodating the very same valency frame shows that the roles accessible to the indirect object do not form a family of interconnected conceptual relations occupying a homogeneous section within a consistent portion of semantic space. This testifies to the conclusion that, although it came into being as a form of expression of a family of allative roles, as shown in the course of the previous section, the indirect object has been desemanticized, generalized, extended — in a word, grammaticalized — as a coding device for different kinds of three-place verbs independently of their content, i.e., as an empty grammatical relation.

162 Chiara Fedriani and Michele Prandi

Evidence for the completion of the formal reshaping of the Dative role complex comes from the fact that in all modern Romance languages11 the prepositional strategy is also extended to a wide and heterogeneous family of margins, including Possessor, Beneficiary, Maleficiary, Substitute for the Agent, and Involuntary Causer, sometimes referred to in the literature as Involuntary Agent (cf. Haspelmath 1993: 291–3, Kittilä 2005, Næss 2007: 82): (37) Maria ha lavato i vestiti a Paolo (Beneficiary + Substistute of the Agent + Possessor) Maria le lavaba la ropa a Pablo Marie a lavé à Paul ses vêtements ‘Mary washed the clothes to/for Paul’ (38) A Maria è bruciata la torta (Involuntary Agent + Maleficiary + Possessor) A Maria se le quemó la tarta ‘Mary (unintentionally) burned the cake’

This extension was not motivated by the allative content of the preposition a/à, as one would have expected within a punctual coding regime, but by a transfer of the whole construction subject — object — indirect object to a family of two-place verbs. A two-place verb like wash, for instance, receives an indirect object as if it were a three-place verb (ex. 37). The essential meaning of the extension, from a relational coding form of a grammatical relation to a punctual coding form of a conceptual relation, already marks an interesting point, if it is true that every time the diachronic development of these markers can be traced back in time, this testifies to an extension from Beneficiary to Recipient (Luraghi 2014; see also Heine et al. 1991). However, what deserves special attention is the way the extension takes place. Let us first consider example (39), where the indirect object can express a Beneficiary, a Substitute for the Agent, or both: (39) Maria ha cucinato una torta a Paolo Mary bake.PAST.3SG ART.F.SG cake to Paul ‘Mary baked a cake to Paul’; ‘Mary baked a cake for Paul/instead of Paul’

The coding form is certainly not motivated by the preposition, as it is, for instance, the standard form of the Beneficiary in Italian with the preposition per ‘for’ (Luraghi 2014). The allative meaning of the preposition a ‘to’ could perhaps motivate the

11.  To be precise, it has to be remarked that the construction is lost in Modern Romanian. However, Nørgård-Sørensen, Heltoft & Schøsler (2011: 255) discuss this point and, following Densusianu (1938: 144), warn that the prepositional schema at first also spread to Romanian, as we have traces of ad and the Accusative in the earliest texts and still in colloquial Romanian.



Exploring a diachronic (re)cycle of roles 163

expression of the Beneficiary, whose content is compatible with motion, but not of the Substitute for the Agent, which has nothing semantically allative in it. A hypothesis that has been put forward on this point is that the role of Recipient has been transferred onto the target process together with the indirect object and reinterpreted as a Beneficiary (see, for instance, Goldberg 1995). This explication is correct in that it identifies the source of the extension with the transfer of the whole construction. However, several objections can be raised to this proposition. First, it rests on two presuppositions that are not self-evident, namely: i. the idea that the indirect object is in any given case the expression of the Recipient or of some metaphorically connected role, and ii. the assumption that there is an intrinsic conceptual continuity between Recipient and Beneficiary. Now, the former presupposition is challenged by the circumstance that the indirect object is a grammatical relation compatible with many different roles and not bound to any, so that its transfer does not imply the transfer of a Recipient; the latter is challenged by the fact that, however able to admit a beneficial implication under some circumstances, a Recipient is not comparable with a Beneficiary. A Recipient is the third argument of a transaction, whereas a Beneficiary is a peripheral role potentially compatible with any kind of action. Moreover, even if it were fit for the Beneficiary, the explication under scrutiny would not account for such peripheral roles as Substitute for the Agent, External Possessor, or Involuntary Causer,12 which have nothing in common with Recipient. We therefore claim that a different process has taken place. While the indirect object is transferred onto a two-place verb as a grammatical relation devoid of an intrinsic content, a two-place verb (for instance, cucinare ‘to cook’) provides no argument to fit this void position. The outcome is that the void position is filled by inference with a peripheral role, or, more typically, with an accumulation of peripheral roles consistent with the process, as illustrated by examples (36) and (37) above, where the only criterion is conceptual consistency with the two-place process. These constructions are interesting for two reasons. First, they document the conversion of a relational coding form into a punctual one, which requires the intervention of some kind of conceptual motivation. Moreover, in this case the conceptual motivation is not provided by the coded meaning of the preposition, which is not reactivated even in the most favourable cases, but by consistent 12.  Goldberg’s explanation is influenced by the circumstance that within the English ditransitive construction similar to the Romance one — John baked Mary a cake — the indirect object is only compatible with the Beneficiary and excludes any other peripheral role. However, this is not the case in the Romance languages (see Prandi 2008).

164 Chiara Fedriani and Michele Prandi

inferences based on the connected conceptual contents. Although disregarded in the current relevant literature, this point should not be a surprise: along with the coded meaning of linking words, inference is one of the roots of conceptual motivation active in punctual coding. To conclude, metaphorically motivated (and thus highly semantically specialized) extensions of ad ‘to’ in Latin gradually paved the way for a process of generalization. As a result, ad/a/à ‘to’ underwent semantic bleaching and grammaticalized, thus proving to be more prone to take on new syntactic functions, notably the grammatical relation of indirect object (see, e.g., Selig 1991: 198, Fagard 2010: 114, Adams 2011: 271). Modern French has been claimed to be the most grammaticalized among the Romance languages in this respect, à having taken over a wider range of functions — for instance, some temporal and instrumental ones, which are less frequent in modern Italian and not attested in Spanish — and having a more structural than representational role (Goyens, Lamiroy & Melis 2002); or, as we would say, a relational rather than a punctual status. 6. Conclusions In the course of this chapter we have tackled some issues pertaining to the diachronic substitution of the Dative role complex from Latin to Romance. We have depicted this process as a sort of cycle whereby an earlier form of coding has gradually been replaced by another, functionally equivalent strategy. We have provided evidence for the claim that this replacement first took place under a punctual coding regime along the lines of a semantically constrained path and discussed the relative speed with which the diachronic development of multifunctional coding devices can evolve and what semantic factors can determine such a shift and influence its course. In Latin, roles such as Recipient, Addressee, and Goal of three-place verbs as well as some second arguments of semantically low-transitive two-place verbs were expressed by the Dative case, which was also used to express margins in some contexts (Section 3). However, as early as in Plautus, a competing pattern featuring the preposition ad with the Accusative started to gain ground over the Dative with a restricted set of trivalent verbs whose third argument could be conceptualized as a Goal, thus enhancing the extension of the motion valency scheme. In Section 4, we explored in detail the dynamics through which the Dative role complex began to be “recycled” by the analytical prepositional strategy. First, we showed how the gradual spread of the prepositional strategy is diachronically motivated by a metaphorical projection of the Goal onto the Recipient and the Addressee, which are the roles entrusted to the indirect object in the



Exploring a diachronic (re)cycle of roles 165

presence of verbs of giving and of communication (Section 3). All of the verbs that initially admit the motion schema involve as an integral part of their semantics a path along which an entity travels. Second, we provided evidence for a principled distinction between in absentia and in praesentia transfers as a compelling motivation for the selection of the prepositional pattern. While the former transfers imply a concrete motion across space in order for the (metaphorical) Goal to be reached, thus facilitating the projection of the motion valency frame, the latter are rather characterized by manual delivery and therefore do not entail any kind of displacement, being more forcefully attached to the maintenance of the Dative. Since they do not provide a clear semantic motivation for the choice of the prepositional schema, they less frequently admit it until Late Latin. The data have revealed that this alternative argument realization only takes place in the domain of punctual coding and is clearly semantically motivated, as the partial substitution of the Dative case by ad with the Accusative is governed by a principle involving the decreasing importance of the idea of motion and an increasing cognitive cost of the recategorization of the third argument as a metaphorical Goal. In Latin, the substitution is limited to cases where the idea of motion is relevant and transparent, and relational forms of coding are therefore excluded from this process of extension. This is why two-place verbs and margins which do not adhere to the model of an allative transaction only admit encoding with the Dative through Late Latin. By the Early Romance language stage, things have changed: the collapse of case morphology leads to an immediate extension of the innovative form of indirect object to cover all the roles belonging to the Dative complex and beyond. As a result, the prepositional strategy spreads to verbs which do not imply motion from a Source to a Goal, thus becoming open to roles other than the metaphorical goal of a transaction. Parallel to that, the preposition underwent a process of semantic bleaching and grammaticalized as a functional element entitled to express a void grammatical relation. From a theoretical perspective, several points have emerged from the foregoing discussion. First, it has to be stressed that in both complexes the roles encoded in a relational regime in the presence of grammatical relations do not share a homogeneous conceptual core, since roles whose allative content has historically motivated the emergence and expansion of the prepositional turn coexist with roles whose coding form eludes any transparent semantic motivation (cf., e.g., Latin mitto ‘I send’ vs. nego ‘I deny’ and modern Italian dare ‘to give’ and rubare ‘to steal’, which even projects onto the indirect object the Source instead of the “expected” Destination). Second, in the case of the indirect object, the change of form entails two successive and opposite changes of coding regime: from a relational regime (the Latin

166 Chiara Fedriani and Michele Prandi

Dative) to a punctual one (ad with the Accusative), and from a punctual regime again to a relational one (a/à + NP as expression of indirect object in Romance). In this way, the Dative cycle confirms the relevance of the distinction between a relational and a punctual coding regime, which in turn justifies both the weight of conceptual motivation, both synchronically and diachronically, and the room left for arbitrary encoding synchronically. Third, the conversion of a punctual coding form into a relational one can be considered an instance of grammaticalization, which affects cases as well as prepositional phrases. A semantic case may become a structural case: such a functional switch probably took place during the prehistory of the Latin Dative, which passed from a punctual coding form of the Recipient to a relational coding form of the grammatical relation of indirect object, compatible with a heterogeneous constellation of roles. In the same way, a prepositional phrase may develop from the direct expression of a semantically transparent relation to encoding a grammatical relation compatible with many different roles. This was certainly the path of the form ad/a/à + NP within the history of Latin and during the development of the Romance languages. Starting as a punctual form of expression of the Goal of a motion, the expression has gradually become the general form of expression of the void grammatical relation of the indirect object. After its gradual extension to processes either metonymically connected with motion or metaphorically redescribed as kinds of motion, it gradually grammaticalized, and could express a broad family of roles devoid of any allative content by the dawn of the Romance languages. With regard to margins, the most interesting point concerns the way one and the same coding form shifts from a relational coding regime towards a punctual one — from the encoding of the indirect object to the expression of a heterogeneous set of margins — which in turn requires the activation of some form of conceptual motivation. If we look for the conceptual reasons of this development, however, we realize that they are not the same in Romance languages as in Latin. The use of Dative to encode peripheral roles in Latin was probably motivated by the tendency of the case to encode non-agentive and low-affected animate referents only loosely involved in the event, which probably represented a trace of the former conceptual motivation of the case. If we look at the set of peripheral roles encoded by the form a/à + NP in the Romance languages, it is clear that the conceptual motivation does not rely on a resemantization of the preposition that had taken over the function of the case and had already been bleached when engaged in the expression of the indirect object, and in particular on a reactivation of its allative meaning. Instead, motivation relies on the other complementary facet of punctual coding expressions, that is, on the capability of their conceptual content to motivate inferences designed to outline the relevant conceptual relations. We know that the constellation of margins including Beneficiary, Maleficiary, Substitute for



Exploring a diachronic (re)cycle of roles 167

the Agent, Involuntary Causer, and External Possessor do not share any allative feature, which alone could have justified a resemantization of the preposition. If they came to be encoded by the prepositional form, it was owing to a transfer of the whole ditransitive construction onto a family of one- or two-place verbs. As the transfer involves a void form, the only way to fill it with a peripheral role is inference, motivated by the connected conceptual contents. The independence of the form from the allative meaning of the preposition accounts for its versatility, and in particular for its aptitude to encode a constellation of roles whose content has nothing allative about it. One final reflection regards the balance that holds between conceptual motivation and arbitrariness. We have seen that the distinction between relational and punctual coding is fully justified synchronically, where the two forms correspond to opposite and complementary criteria and display opposite and complementary properties: relational forms are independent of conceptual motivation, whereas punctual coding forms are at the service of independent conceptual relations, and thus motivated. Although relevant in synchronic terms, however, the division of labor between relational and punctual coding is mobile diachronically, where the two regimes take turns in both directions. On the one hand, relational coding forms can be attacked by punctual competitors: in Latin, for instance, the Dative form of the indirect object is gradually replaced by the semantically transparent alternative ad + Accusative. On the other hand, punctual coding forms may be extended to non-motivated functions to become relational coding forms. The twofold nature of Romance indirect objects, a relational coding form of a void grammatical relation arising out of a punctually motivated coding form of a semantically transparent relation, leads to the reevaluation of the dichotomy between synchrony and diachrony (Saussure 1974[1916]) and its import for linguistic description and, in particular, for the relationship between relational, arbitrary coding and punctual, motivated coding, both of which compete, co-occur, and interact in language and are in some ways equally functional for different reasons.

Acknowledgements This contribution is the result of fruitful discussions and mutually inspiring collaboration between the authors. Chiara Fedriani is responsible for Sections 1 and 3, and Michele Prandi for Sections 2 and 4. The conclusions have been written by both authors together. We would like to thank Pierluigi Cuzzolin, three anonymous reviewers, and the editors of this volume for commenting on earlier versions of this chapter, thereby significantly enhancing its final form.

168 Chiara Fedriani and Michele Prandi

Abbreviations ABL ​= ​ablative; ACC ​= ​accusative; ACT ​= ​active; ART ​= ​article; CONJ ​= ​conjunction; DAT ​= ​dative; F ​= ​ feminine; FUT ​= ​future; GEN ​= ​genitive; GER ​= ​gerund; GRD ​= ​gerundive; IMP ​= ​imperative; IMPF ​= ​ imperfect; INF ​= ​infinitive; INT ​= ​interrogative; M ​= ​masculine; N ​= ​neuter; PART ​= ​participle; PASS ​ = ​passive; PAST ​= ​past; PERIPHR ​= ​periphrastic; PF ​= ​perfect; PL ​= ​plural; POSS ​= ​possessive; PPF ​= ​ pluperfect; PRON ​= ​pronoun; PRS ​= ​present; PTC ​= ​particle; REL ​= ​relative; SG ​= ​singular; SUBJ ​= ​ subjunctive; VOC ​= ​vocative.

References Adams, James N. 2011. Late Latin. In J. Clackson (ed.), The Blackwell Companion to the Latin Language, 257–283. Oxford: Blackwell. Baldi, Philip & Andrea Nuti. 2010. Possession. In P. Cuzzolin & Ph. Baldi (eds.), New Perspectives on Historical Latin Syntax, 239–388. Berlin: Mouton de Gruyter. Baños Baños, José Miguel. 1996. Litteras Neroni / ad Neronem mittere: ¿alternancia dativo / ad + acus.? In A. Agud, J.A. Fernández Delgado & A. Ramos Guerreira (eds.), Las lenguas de corpus y sus problemas lingüísticos, 217–236. Madrid: Ediciones Clásicas. Baños Baños, José Miguel. 1998. Sintaxis y semántica del dativo objeto indirecto: su concurrencia con ad-acus. en latín clásico. In M.E. Torrego (ed.), Nombres y funciones: estudios de sintaxis griega y latina, 11–41. Madrid: U.A.M. Ediciones Clásicas. Baños Baños, José Miguel. 2000. Vulgarismos sintácticos en Plauto (II): quae ad patrem vis nuntiari. In B. García Hernandez (ed.), Latín vulgar y tardío, 1–15. Madrid: Ediciones Clásicas. Baños Baños, José Miguel. 2009. Dativo. In J. M. Baños Baños (ed.), Sintaxis del latín clásico, 185–209. Madrid: Liceus Ediciones. Bastardas Parera, Juan. 1953. Particularidades sintácticas del latín medieval. Barcelona: Escuela de Filología. Bennett, Charles Edwin. 1914. Syntax of Early Latin. Boston: Allyn and Bacon. Blake, Barry J. 2001[1994]. Case, 2nd edn. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9781139164894

Blume, Kerstin. 1998. A contrastive analysis of interaction verbs with dative complements. Linguistics 36 (2). 253–280.  doi: 10.1515/ling.1998.36.2.253 Cennamo, Michela. 1999. Late Latin pleonastic reflexives and the unaccusative hypothesis. Transactions of the Philological Society 97(1). 103–150.  doi: 10.1111/1467-968X.00046 Chomsky, Noam A. 1957. Syntactic Structures. The Hague & Paris: Mouton. Chomsky, Noam. A. 1966. Using Language. Cambridge: Cambridge University Press. Christol, Alain. 1998. Marquage oblique des actants. In J. Feuillet (ed.), Actance et valence dans les langues de l’Europe, 457–523. Berlin: Mouton de Gruyter. Dąbrowska, Ewa. 1997. Cognitive Semantics and the Polish Dative. Berlin: Mouton de Gruyter.  doi: 10.1515/9783110814781 Densusianu, Ovide. 1938. Histoire de la langue roumaine II. Paris: Librairie Ernest Leroux. Dik, Simon. C. 1989[1997]. The Theory of Functional Grammar. Part I: The Structure of the Clause, 2nd revised edn. Berlin & New York: Mouton de Gruyter. Ernout, Alfred & François Thomas. 1953. Syntaxe latine. Paris: Klincksieck.

Exploring a diachronic (re)cycle of roles 169



Fagard, Benjamin. 2010. Espace et grammaticalisation – L’évolution sémantique des prépositions dans les langues romanes. Sarrebruck: Editions Universitaires Européennes. Fried, Mirjam. 2010. Between verb semantics and interpersonal meanings in participant realization. Paper presented at theWorkshop on Variation and Change in Argument Realization, Naples and Capri, May 27-30, 2010. Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Goyens, Michèle, Lamiroy, Béatrice & Ludo Melis. 2002. Déplacement et repositionnement de la préposition à en français. Linguisticae Investigationes 25(2). 275- 310.  doi: 10.1075/li.25.2.06goy

Haiman, John (ed.). 1985. Iconicity in Syntax. Amsterdam: John Benjamins.  doi: 10.1075/tsl.6 Haspelmath, Martin. 1993. A grammar of Lezgian. Berlin: Mouton de Gruyter. Haudry, Jean. 1968. Les emplois doubles du datif et la fonction du datif indo-européen. Bulletin de la Société de Linguistique de Paris 63. 141–159. Heine, Bernd, Ulrike Claudi & Friederike Hünnemeyer. 1991. Grammaticalization: A Conceptual Framework. Chicago: University of Chicago Press. Hofmann, Johann Baptist. 1936. Lateinische Umgangssprache. Winter: Heidelberg. Hofmann, Johann Baptist & Anton Szantyr. 1972 [1965]. Lateinische Syntax und Stilistik. München: Beck (= Leumann-Hoffmann-Szantyr, Lateinische Grammatik, 2nd edn.). Jacob, Daniel. 1991. Dativ im Französischen? Zur Funktionsweise und Semantik der Ergänzungsklasse ‘à + NP’. In P. Koch & Th. Krefeld (eds.), Connexiones Romanicae: Dependenz und Valenz in romanischen Sprachen, 157–186. Tübingen: Niemeyer. Jakobson, Roman. 1966. A la recherche de l’essence du langage. In Problèmes du langage, Diogène 51: 22–38. English translation 1971, “Quest for the Essence of Language”, in Roman Jakobson, Selective writings, vol. 2: Word and Language, 345–359. The Hague/ Paris: Mouton. Kittilä, Seppo. 2005. A typology of involuntary agent constructions. Word 56. 341–81. Kittilä, Seppo. 2006. On distinguishing between ‘recipient’ and ‘beneficiary’ in Finnish. In M. Helasvuo & L. Campbell (eds.), Grammar from the Human Perspective: Case, space and person in Finnish, 129–152. Amsterdam: John Benjamins.  doi: 10.1075/cilt.277.11kit Kittilä, Seppo, Katja Västi & Jussi Ylikoski. 2011. In S. Kittilä, K. Västi & J. Ylikoski (eds.), Introduction to case, animacy and semantic roles. Case, Animacy and Semantic Roles, 1–26. Amsterdam: John Benjamins.  doi: 10.1075/tsl.99.01kit König, Ekkehard & Martin Haspelmath. 1998. Les constructions à possesseur externe dans les langues de l’Europe. In J. Feuillet (ed.), Actance et valence dans les langues de l’Europe, 525–606. Berlin: Mouton de Gruyter. Kühner, Raphael & Carl Stegmann. 1976. Ausführliche Grammatik der lateinischen Sprache. Hannover: Verlag Hahnsche Buchhandlung. Kuryłowicz, Jerzy. 1964. The Inflectional Categories of Indo-European. Heidelberg: Winter. Lakoff, George. 1987. Women, Fire, and Dangerous Things: What Categories Reveal About the Mind. Chicago: Chicago University Press.  doi: 10.7208/chicago/9780226471013.001.0001 Langacker, Ronald. 1992[1991]. Concept, Image and Symbol: The Cognitive Basis of Grammar, 2nd edn. Berlin & New York: Mouton De Gruyter. Lazard, Gilbert. 1998. Actancy. Berlin: Mouton de Gruyter. Lehmann, Christian. 1991. Predicate classes and PARTICIPATION. In H. Seiler & W. Premper (eds.), Partizipation: das sprachliche Erfassen von Sachverhalten, 183–239. Tübingen: Gunter Narr.

170 Chiara Fedriani and Michele Prandi Lehmann, Christian. 2006. Participant roles, thematic roles and syntactic relations. In T. Tsunoda & T. Kageyama (eds.), Voice and Grammatical Relations. Festschrift for Masayoshi Shibatani, 167–190. Amsterdam: John Benjamins. Levin, Beth. 1993. Verb Classes and Alternations: A Preliminary Investigation. Chicago: University of Chicago Press. Löfstedt, Bengt. 1961. Studien über die Sprache der langobardischen Gesetze. Beiträge zur frühmittelalterlichen Latinität. Stockholm: Almqvist & Wiksell. Longacre, Robert E. 2007[1985]. Sentences as combinations of clauses. In T. Shopen (ed.), Language Typology and Syntactic Description, Vol. II: Complex Constructions, 2nd edn., 372–420. Cambridge: Cambridge University Press. Luraghi, Silvia. 2010. Adverbials. In Ph. Baldi & P. Cuzzolin (eds.), New Perspectives on Historical Latin Syntax, vol. 2, 19–96. Berlin: Mouton de Gruyter. Luraghi, Silvia. 2014. Plotting diachronic semantic maps: The role of metaphor. In S. Luraghi & H. Narrog (eds.), Perspectives on Semantic Roles, 101–152. Amsterdam: John Benjamins. Mathieu, Eric. 2006. Quirky subjects in Old French. Studia Linguistica 60(3). 282–312.  doi: 10.1111/j.1467-9582.2006.00128.x

Maiden, Martin. 1996. A Linguistic History of Italian. New York: Longman. Menge, Hermann. 2007. Lehrbuch der lateinischen Syntax und Semantik, völlig neu bearbeitet von Thorsten Burkard und Markus Schauer, 3. Auflage. Darmstadt: Wissenschaftliche Buchgesellschaft. Molinelli, Piera. 1998. Dai casi alle preposizioni in latino: analisi sociolinguistica e spiegazione tipologica. In P. Ramat & E. Roma (eds.), Sintassi storica, 147–166. Roma: Bulzoni. Næss, Åshild. 2007. Prototypical Transitivity. Amsterdam: John Benjamins.  doi: 10.1075/tsl.72 Newman, John. 1996. Give: A Cognitive Linguistic Study. Berlin & New York: Mouton de Gruyter. Nørgård-Sørensen, Jens, Lars Heltoft & Lene Schøsler. 2011. Connecting Grammaticalisation. Amsterdam: John Benjamins.  doi: 10.1075/sfsl.65 Palmer, Frank Robert. 1994. Grammatical Roles and Relations. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9781139166638 Pinkster, Harm. 1990. The development of cases and adpositions in Latin. In H. Pinkster & I. Genee (eds.), Unity in Diversity, 195–209. Dordrecht: Foris. Prandi, Michele. 2004. The Building Blocks of Meaning: Ideas for a Philosophical Grammar. Amsterdam: John Benjamins.  doi: 10.1075/hcp.13 Prandi, Michele. 2008. Transferring constructions: Grammatical relations and roles. In M. Papi Bertuccelli, A. Bertacca & S. Bruti (eds.), Threads in the Complex Fabric of Language: Linguistic and Literary Studies in Honour of Lavinia Merlini Barbaresi, 293–305. Pisa: Felici Editore. Reddy, Michael J. 1993. The conduit metaphor: A case of frame conflict in our language about language. In A. Ortony (ed.), Metaphor and Thought, 164–201. Cambridge: Cambridge University Press.  doi: 10.1017/CBO9781139173865.012 Saussure, Ferdinand de. 1974[1916]. Cours de linguistique générale. Paris: Payot. Critical edition by T. de Mauro, Paris: Payot, 1972. Engl. transl.: Course in General Linguistics. London: Fontana / Collins. Selig, Maria. 1991. Inhaltskonturen des ‘Dativs’: Zur Ablösung des lateinischen Dativs durch ad und zur differentiellen Objektmarkierung. In P. Koch & Th. Krefeld (eds.), Connexiones Romanicae: Dependenz und Valenz in romanischen Sprachen, 187–211. Tübingen: Niemeyer. Serbat, Guy. 1996. Grammaire fondamentale du latin. Tome IV, L’emploi des cas en latin, Vol. 1. Louvain et Paris: Peeters.



Exploring a diachronic (re)cycle of roles 171

Steinitz, Renate. 1969. Adverbial-Syntax. Berlin: Akademie Verlag. Sznajder, Lyliane. 2012. Dixit autem serpens ad mulierem / Mulieri quoque dixit: la double expression de l’allocutaire dans les propositions introductrices de discours directs dans la Vulgate. F. Biville, M. Lhommé & D. Vallat (eds.), LATIN VULGAIRE-LATIN TARDIF IX, Actes du IXe colloque international ‘latin vulgaire – latin tardif,’ (Lyon, 2-6 septembre 2009), 271–290. Lyon: Collection de la maison de l’Orient et de la Méditerranée 49, Série Linguistique et Philologie 8. Théoret, Michel.1982. Les discours de Cicéron: La concurrence du tour casuel et du tour prépositionnel. Montreal: Presse de l’Université de Montréal. Väänänen, Veikko. 1963. Introduction au latin vulgaire. Paris: Klincksieck. Van Hoecke, Willy. 1996. The Latin dative. In W. Van Belle & W. Van Langendonck (eds.), The Dative, Vol. I: Descriptive Studies, 3–37. Amsterdam: John Benjamins. Van Langendonck, Willy. 1998. The dative in Latin and the indirect object in Dutch. In W. Van Belle & W. Van Langendonck (eds.), The Dative, Vol.II: Theoretical and Contrastive Studies, 221–259. Amsterdam: John Benjamins. Villar, Francisco. 1981. Dativo y locativo en el singular de la flexión nominal indoeuropea. Salamanca: Universidad de Salamanca. Woodcock, Eric Charles. 1959. A New Latin Syntax. London: Methuen. Zamboni, Alberto. 2000. Alle origini dell’italiano. Dinamiche e tipologie della transizione dal latino. Roma: Carocci.

Functive phrases in typological and diachronic perspective Denis Creissels

University of Lyon

This chapter is concerned with a particular semantic type of noun phrases in secondary predicate function, illustrated by as a negotiator in He was sent as a negotiator. It explores the cross-linguistic regularities in the polysemy patterns characteristic of the markers and constructions having the expression of this meaning as one of their possible uses, and discusses the grammaticalization paths in which they are involved. Keywords: functive, essive, grammaticalization, secondary predicate, polysemy

1. Introduction The combination of a noun phrase or adposition phrase with a verb is not always designed to express a participant in an event encoded by the verb. In addition to participative roles (such as agent, patient, instrument, beneficiary, etc.), noun phrases or adposition phrases may also fulfill circumstantial roles, in which they refer to circumstances of the event (place, time, manner, cause, etc.), and predicative roles, in which they express secondary predications about participants. Nouns may also combine with verbs into complex predicates of the type commonly termed light-verb constructions, in which they contribute to the definition of a type of event. This chapter deals with a particular role that noun phrases or adposition phrases may fulfill in the construction of verbs, for which the label functive is used. As a negotiator in He was sent as a negotiator is a typical example of a functive phrase. Crucially, a negotiator does not refer here to an additional participant in the event described by He was sent. In this construction, as a negotiator expresses a predication about the referent of the subject (he), and the semantic contribution of this phrase can be paraphrased as to fulfill the role of a negotiator.

doi 10.1075/bct.88.07cre 2016 © John Benjamins Publishing Company

174 Denis Creissels

The central concern of this chapter is the diachronic explanation of the crosslinguistic variation observed in the polysemy of markers or constructions having the expression of the functive as one of their possible uses. After some preliminary remarks aimed at avoiding possible misunderstandings about the scope of this study and its basic assumptions (Section 2), I present cross-linguistic data on the syncretisms in which functive markers are involved (Sections 3 to 6), and I discuss the grammaticalization paths that may lead to the polysemy patterns observed (Sections 7 to 9). The obvious difficulty one has to face in such a study is that most grammatical traditions do not include the functive among the notions whose involvement in the structuring of languages must be systematically taken into account. Consequently, gathering data representative of the cross-linguistic diversity in the encoding of this semantic role and its possible connections with other roles cannot be achieved by simply consulting the indexes of reference grammars, and this practically excludes a rigorous typological study on the basis of a representative sample of languages. The sample of languages I have used is certainly biased by my familiarity with a small number of language families, and the generalizations proposed in this chapter must be viewed as tentative.1 2. Definition and first observations 2.1 Definition In this chapter, the term ‘functive’ is used with the following definition:

1.  The following persons have contributed to this chapter by providing me with data, or helping me to orient myself in questions I am not familiar with: Ayten Babaliyeva (Tabassaran, Lezgi, Azeri), Mehmet-Ali Akıncı (Turkish), Sokhna Ba-Diop (Wolof), Alain-Christian Bassène (Jóola Banjal), Bertrand Boiron (Hungarian), Nthatisi Bulane (Southern Sotho), Jean-François Chaléat (Japanese), Claudine Chamorreau (Purépecha), Hilary Chappell (Chinese dialects), Injoo ChoiJonin (Korean), Michael Daniel (Archi), Danh Thành Do-Hurinville (Vietnamese), Gérard Dumestre (Bambara), Dmitri Ganenkov (Dargi), Csilla Kász (Arabic), Victoria Khurshudian (Eastern Armenian), Jérémie Kouadio (Baule), Joseba Lacarra (Basque), René Lacroix (Laz), Timur Maisak (Agul), Solmaz Merdanova (Agul), Zarina Molochieva (Chechen), Céline Mounole (Basque), Claire Moyse (Xaracuu, Wallisian, Futunan), Irina Nikolaeva (Nenets), Hayat Omar (Amharic, Oromo), Daniel Petit (Indo-European), Françoise Rose (Émérillon, Mojeño Trinatario, and other Amazonian languages), Wu Tong (Mandarin Chinese), Manana Topadze (Georgian), Nicolas Tournadre (Tibetan), Martine Vanhove (Maltese). This version has also benefited from Ekkehard König and Wolfgang Schulze’s comments on previous versions, and (last but not least) from the comments of two anonymous reviewers.



Functive phrases in typological and diachronic perspective 175

1. A functive phrase is either a noun phrase headed by a noun N normally used to refer to concrete entities (animate beings or concrete things), or the combination of such a noun phrase with an adposition. 2. A functive phrase is syntactically a direct dependent of the verbal head of the clause. 3. A functive phrase does not refer to a participant identifiable as an N and distinct from the participants referred to by the other noun phrases in the construction of the same verb, but predicates the property of fulfilling the role of an N, taking one of the participants as its argument. The noun phrase about whose referent the functive phrase predicates the property of fulfilling the role of an N can be designated as the controller of the functive phrase. For example, in I am talking to you as your friend, I is the controller of the functive phrase as your friend. Functive phrases are sometimes referred to as role phrases, for example, by Haspelmath & Buchholz (1998), whose definition (“Role phrases express the role or function in which a participant appears”) is equivalent to the definition of the functive formulated above. These authors also use the term of functive (which as far as I know has been used for the first time with reference to role phrases in Haspelmath’s grammar of Lezgian), but they restrict its use to languages in which nouns are inflected for case and one of the case forms of nouns is used to express this notion. In English, noun phrases in functive role are typically introduced by the preposition as — Ex. (1). (1)

English: as-phrases in functive role a. John is working as a teacher. b. I got these books as a gift. c. I mentioned this as an example. d. I am talking to you as your friend. e. We were given melon with prosciutto as a first course. f. I used my sweater as a pillow.

The expression of the functive as defined above is however not the only possible use of ‘as + NP’ sequences. Ex. (2) illustrates uses of as that the adopted definition prevents from being recognized as functive, since they do not really involve the notion of function or role, although their semantic closeness to the functive is obvious.

176 Denis Creissels

(2)

English: as-phrases in roles distinct from the functive a. She appeared on stage as a man. (participant-oriented similative) b. As a child he was very rebellious. (temporally-bound equative)2 c. As a woman you should not tell that. (individual-level equative)

The syncretisms involving markers or constructions that have the expression of the functive role as one of their possible uses show important cross-linguistic variation. For example, in Mandinka, the postposition tí marking the functive phrase in (3a) cannot be used to mark nouns expressing temporally-bound identification, which are marked by the suffix -mâa also used in the expression of participantoriented similative — Ex. (3b–c). Tí is not used with secondary predicates expressing individual-level identification either, and Mandinka has a special suffix -tôo for this function — Ex. (3d). By contrast, transformative, which cannot be expressed in English by means of as, is one of the possible uses of the Mandinka postposition tí — Ex. (3e–f), and tí also occurs in independent equative clauses — Ex. (3g).3 (3) Mandinka (p.d.) a. Ŋ́ ŋá até le lóŋ ŋ́ faamáa ti . 1sg cmp.pos 3sg:emph foc know 1sg father:def obl ‘It is him that I knew as my father .’ (functive) ́ b. Ŋ ŋá í faamâa díndím-mâa  lóŋ. 1sg cmp.pos 2sg father:def child-spr know ‘I knew your father as a child .’ (stage-level equative) c. Íbúlíisá saa-máa  naatá Awá kaŋ. Satan snake-spr come:cmp.pos Eve on ‘Satan came to Eve as a snake.’ (participant-oriented similative) d. Íte musu-tôo  mâŋ ñánna ñiŋ fóla. 2sg woman-spr cmp.neg must dem tell:inf ‘ As a woman  you should not tell that.’ (individual level identification)

2.  Equative is taken here in its broad meaning, conflating equative stricto sensu with specificational and identificational — for a recent discussion of these notions within the frame of a typology of copular clauses, see Mikkelsen (2011). Note that this use of equative has nothing to do with the label equative case found in the description of the nominal inflection of Sumerian and a few other languages. In the terminology used in the present paper, such case forms would rather be labeled similative. 3.  In the glosses of the Mandinka examples, the suffixes -tôo and -mâa are glossed SPR (secondary predicate marker), because they are used exclusively with nouns in secondary predicate function, whereas the postposition tí is simply glossed OBL (oblique NP marker), because in addition to its use with NPs in predicate function, it also occurs with NPs expressing a variety of participative roles — see 5.1.



Functive phrases in typological and diachronic perspective 177

e. Jíyo yelematá dolóo ti . water:def change:cmp.pos wine:def obl ‘The water changed into wine .’ (transformative) f. Kewólu ye yíríjúwo lésé kúlúŋo ti . man:def:pl cmp.pos tree_trunk:def carve boat:def obl ‘The men carved the tree trunk into a boat .’ (transformative) g. Ñiŋ yíróo mu bantáŋo le ti . dem tree:def cop silk_cotton_tree:def foc obl ‘This tree is a silk-cotton tree .’ (independent clause  expressing identification)

Moreover, the Mandinka postposition tí is not limited to the encoding of predicative roles. It is also fully productive as a marker of the standard of comparison in the construction expressing the comparative of superiority — Ex. (4a), and is marginally found as a marker of phrases in other non-predicative roles: comitative — Ex. (4b) — and substitutive (‘in exchange for’) — Ex. (4c). (4) Mandinka (p.d.) a. Sánóo le kúlíyáatá kódóo ti. gold:def foc be_heavy:cmp.pos silver:def obl ‘Gold is heavier than silver.’ b. Kewó naata kídóo ti. man:def come:cmp.pos gun:def obl ‘The man brought a gun.’ (lit. ‘came with a gun’) c. I yé joŋólu saŋ kídímúŋkóo ti. 3pl cmp.pos slave:def:pl trade gun:powder:def obl ‘They traded slaves for gun powder.’

2.2 Functive and equative The main difficulty in identifying functive phrases is to draw a distinction between phrases predicating the property of fulfilling the role of an N (N a noun) and phrases predicating other semantic types of identification. The view adopted here is that prototypical equative (the term equative being taken in its broad meaning — see footnote 2) can be defined as the expression of individual-level identification, i.e. the identification of an entity with reference to relatively stable properties, whereas the functive can be defined as the expression of functional identification, or role-bound identification, and therefore constitutes one of the non-prototypical varieties of the equative alongside, for example, temporally-bound identification. A definition of the relationship between functive and equative in terms of relative closeness to prototypes is made necessary by the absence of a clear-cut

178 Denis Creissels

boundary between prototypical and non-prototypical identification, and between the various non-prototypical subtypes of identification, as evidenced by the crosslinguistic variation in the cut-off points between constructions expressing prototypical identification and constructions more or less specialized in the expression of non-prototypical varieties of identification. For example, the closeness between role-bound identification (functive) and temporally-bound identification is obvious, as reflected by the use of the same marker for functive phrases and phrases expressing temporally-bound identification in English and other languages. At the same time, however, Ex. (3) above shows that the distinction is relevant at least in some languages. The sharpness of the distinction between the functive and other varieties of identification crucially depends on the lexical meaning of the noun in functive role. For example, in I used my sweater as a pillow, it is clear that the relationship between sweater and pillow can only be interpreted in terms of functional identification: a sweater is not a pillow, even when used as a pillow. By contrast, in John is working as a teacher, the distinction between functive and individual-level equative tends to blur. Working as a teacher is not exactly the same thing as being a teacher, since being a teacher suggests a permanent state requiring a qualification, whereas working as a teacher suggests that the position is temporary, and does not necessarily correspond to the main qualification of the person that occupies it. Speakers are aware of the distinction, and it is not difficult to imagine contexts in which it is relevant, but in most contexts, John is working as a teacher and John is a teacher can be used interchangeably, since a person who is a teacher (i.e., who has a professional qualification in teaching) is normally expected to work as a teacher, and vice-versa. Reanalysis processes in such transitional contexts are responsible for the fact that, when a language has several constructions or markers expressing various types of identification, the markers used in prototypical functive situations do not necessarily extend their use to all types of identification for which a functive interpretation is conceivable, and at the same time may also be used to encode types of identification for which a functive interpretation is clearly ruled out. 2.3 Functive and transformative The label transformative is used here for a comparative concept defined as follows: noun phrases in transformative role refer to the final state of a participant undergoing a transformation, as in He decided to become a writer, or Jesus made water into wine. Like noun phrases in functive role, noun phrases in transformative role do not refer to participants distinct from those encoded by the other noun phrases



Functive phrases in typological and diachronic perspective 179

included in the same construction, but rather contribute to the expression of a predication about the referent of another noun phrase. The relationship between equative and transformative is obvious, since transformative phrases can be viewed as equative phrases in the scope of a verb encoding a transformation event, and the transformative can therefore be considered a contextual variant of the equative. The closeness between transformative and functive is also obvious, since there is no clear-cut distinction between a change that affects the essential properties of an entity, and the acquisition of a more or less temporary role that does not affect the very nature of the entity concerned. It is therefore not surprising that the constructions or markers found in contexts in which they are unambiguously interpreted as functive are also commonly found in uses that cannot be easily categorized as functive or transformative. We will return to functive-transformative syncretism in Section 3.4. 2.4 Functive phrases in copulative constructions, functive arguments, and functive adjuncts Finnish illustrates the situation in which functive-marked noun phrases (in Finnish, nouns phrases in the Essive case) can be found in copulative constructions in which they contrast with another form of the noun phrase (in Finnish, the Nominative case) used in the same context to express permanent identification. (5) Finnish (Wikipedia: Suomen kieli ulkomaalaisille/Sisältö/Essiivi) a. Hän on opettaja-na. 3sg be.prs.3sg teacher-ess ‘He is teaching (temporarily).’ or ‘He is working as a teacher for the time being.’ b. Hän on opettaja. 3sg be.prs.3sg teacher ‘He is a teacher.’ (This is his profession and is likely to remain so.)

However, this is not a common situation. In most languages, the functive markers used with noun phrases expressing secondary predications in combination with non-copulative verbs cannot be found in copulative constructions, and the meaning unambiguously expressed in Finnish by the construction illustrated in (9a) can only appear as a contextual interpretation of a copulative construction whose default meaning is plain identification, or be specified by the combination of a functive phrase with a verb such as English serve in serve as a clerk, serve as a reminder. Some languages have constructions that can be described as involving a copulative-like use of a verb do combined with a functive noun phrase. The verbs

180 Denis Creissels

found in such constructions are bivalent verbs usually found as the equivalent of English do, with a second argument denoting an activity (do the washing), the entity affected by an activity (do the garden), or the result of an activity (do a lot of damage). In particular, ‘do’ verbs with abstract nouns derived from nouns as their second argument constitute in many languages the usual equivalent of English work / serve as: Agul (East Caucasian) dijark̄a-ʕʷel aq’as lit. ‘do milkmaid-hood’ > ‘work as a milkmaid’, Karata (East-Caucasian) ušter-ɬer gāɬa lit. ‘do teacher-hood’ > ‘work as a teacher’. Such constructions do not count as functive constructions according to the definition adopted here, since the second argument of ‘do’ is an abstract noun denoting an activity. But in some languages, such verbs also have the ability to assign the functive role to nouns normally used to refer to concrete entities; the nouns in question may be encoded in the same way as patients of typical transitive verbs — Ex. (6), or as oblique arguments — Ex. (7). (6)

Japanese (Irène Tamba, p.c.) Kare wa hoogakkoo no kyooshi o shite iru. 3sg top primary_school gen teacher acc do:cvb be ‘He is working as a schoolteacher.’ lit. ‘He is doing schoolteacher.’

(7)

Spanish (p.k.) Mi hermana hace de secretaria. my sister do:prs:3sg de secretary ‘My sister works as a secretary.’ lit. ‘My sister does of secretary.’

As regards functive phrases combined with non-copulative verbs, the definition formulated in Section 2.1 says nothing about their status with respect to the distinction between arguments and adjuncts. However, unambiguous functive constructions in which the functive phrase is an argument are cross-linguistically uncommon. It is true that in many contexts, the distinction between functive and other subtypes of equative, or between functive and transformative, is more or less blurred. But once the importance of these distinctions has been acknowledged, a closer look at constructions including predicative arguments whose marking may suggest a functive interpretation leads to the conclusion that the role of the predicative argument is almost always best analyzed as plain equative or transformative. For example, the construction illustrated by John considers Mary (as) a strong woman is a raising construction, and the embedded predication in the logical structure of this sentence is plain equative rather than functive, as suggested by the paraphrase John considers that Mary is a strong woman. The same applies to Mary in John took Mary for his sister, paraphrasable as John got the (false) impression that Mary was his sister. Functive phrases arguably analyzable as arguments rather than adjuncts will however be examined in Section 2.10.

Functive phrases in typological and diachronic perspective 181



2.5 Functive periphrases From a purely onomasiological perspective, translational equivalents of the functive constructions analyzed in the present paper must exist in all languages, but fully grammaticalized constructions falling under the definition of functive formulated in Section 2.1 do not necessarily exist in all languages. Not all noun phrases that in some way or another express a role fulfilled by the referent of another noun phrase fall within the scope of the present study. The distinction between grammaticalized functive marking and functive periphrases may however be problematic, since expressions of the types discussed in this section often show evidence of grammaticalization. Diachronically, they are a possible source of fully grammaticalized functive marking, and there is no universal criterion making it possible to decide at what point in the grammaticalization process a functive marker has emerged from a construction initially analyzable as a functive periphrasis. A first case is that of formulations such as by way of N, in one’s capacity/nature/ quality of N, in which the notion of role or function is implied by the lexical meaning of the abstract noun that constitutes the head of the construction. Ex. (8) & (9) illustrate the case of languages in which such constructions constitute the usual equivalent of as-constructions in English. (8)

Hindi (Annie Montaut, p.c.) Usne do sal maastar ke rûp men kâm kiyâ hai. 3sg:erg two year schoolteacher gen way in work do pfv ‘He worked two years as a schoolteacher.’

(9)

Wallisian (Claire Moyse, p.c.) ‘E talaga ia ia i tona ‘uhiga ‘aliki. npst speak abs 3sg obl poss.3sg nature4 chief ‘He is speaking as a chief.’

Similarly, in Persian, the meanings expressed in other languages by constructions that count as functive constructions according to the definition adopted here are expressed by means of formulations like be envân ‘with the title of ’, be semat ‘with the mark of ’, or be manzele ‘with the rank of ’. See also Šarić (2008) on the use of u svojstvu / ulozi / funkciji ‘in the quality / role / function’ in Croatian. In some languages, the usual translational equivalent of English ‘as N’ may also involve the derivation of abstract nouns, lit. in N-hood or doing N-hood. Such

4.  Depending on the context, ‘uhiga can variously be translated as ‘reason’, ‘meaning’, ‘nature’, or ‘identity’.

182 Denis Creissels

formulations constitute another possible source of functive markers (see Sections 7.1 & 8.2), but are not counted here as functive constructions proper. Another possibility is the use of bi-verbal constructions implying argument sharing with the following division of labor: the shared argument is expressed as an argument of the verb that assigns it a participant role, whereas the noun phrase in functive role is introduced as an argument of the other verb. As illustrated by Ex. (10), in Japanese, to shite taken as a whole expresses a functive or equative meaning similar to that expressed in English by as, but grammatically, to is a (multifunctional) case particle and shite is the converbial form of suru ‘do’, combined with sumu ‘inhabit’ fulfilling the function of main predicate. (10)

Japanese (Kuwae 1989: 481) Wakai koro gakusei to shite Tookyoo ni sannen sunde itta. young when student to do:cvb Tokyo in three_years inhabit:cvb be:pst ‘I lived three years in Tokyo as a student when I was young.’5

It can however be argued that this decomposition is no longer relevant to syntax, and that, in present-day Japanese, to shite must be considered as a compound marker, in spite of its transparent etymology (Irène Tamba, p.c.). Here again, it is very difficult to decide at what point in its grammaticalization process a periphrasis of this type must be considered as having been converted into a grammaticalized functive construction. Ex. (11) illustrates a similar construction in Karata (East Caucasian), with a converbial form of gāɬa ‘do’ introducing a functive noun phrase that, semantically, expresses a predication about the Nominative argument of k’use idja ‘is sitting’. (11)

Karata (Magomedova & Xalidova 2001) Hĩc̄’u ela q̄arawul gē χʷaj idja k’use. door at guard do:cvb dog cop sit:cvb ‘A dog is sitting at the door as a guard.’

In other languages, similar periphrases are rather of the type commonly designated as serial verb construction. For example, Mandarin Chinese uses serial constructions involving dāng ‘act as’ or zuò ‘do’ — Ex. (12), and similar constructions are found in other Chinese varieties. (12)

Chinese (Wu Tong, p.c.) Ta xuan wo dang hezuo huoban. 3sg choose 1sg act_as collaboration partner ‘He chose me as (his) collaborator.’

5.  The first person subject in the translation of this Japanese sentence is supposed to have been retrieved from the context. The sentence itself provides no clue to identify the person referred to.



Functive phrases in typological and diachronic perspective 183

Finally, as illustrated by Ex. (13), some languages have functive periphrases in which the functive phrase is introduced by a dependent form of a copula, i.e., a word whose literal meaning can be glossed as ‘being’.6 (13)

Turkish (Göksel & Kerslake: 216) Bun-u san-a avukat ol-arak değil, arkadaş ol-arak söyl-üyor-um. dem-acc 2sg-dat lawyer be-cvb neg friend be-cvb say-prog-1sg ‘I’m saying this to you not as a lawyer but as a friend.’

A similar use of a dependent form of a copula is found in Amharic (Ethiosemitic), Oromo (Cushitic), Sotho-Tswana (Bantu), etc. Not surprisingly, this type of formulation is particularly common with nouns whose lexical meaning blurs the distinction between functive and prototypical equative. 2.6 Functive phrases in the typological literature From the point of view of a general typology of secondary predicates, functive phrases in non-copulative constructions are a subtype of depictives, both syntactically and semantically: syntactically, depictives as defined by Schultze-Berndt & Himmelman (2004) may not only be noun phrases or adposition phrases, but also adjectives, or phrases headed by non-finite verb forms, and semantically, the definition of depictives does not specify the nature of the predication they express, whereas the definition of the functive posited above limits this notion to noun phrases or adposition phrases paraphrasable as ‘fulfilling the role of an N’. Since functive phrases in non-copulative constructions are a particular type of secondary predicate, more or less developed discussions about functive phrases in the sense of the definition put forward here can be found in a number of studies devoted to secondary predication, either on particular languages or in a more general perspective — see among others Nichols (1978), Williams (1980), Pinkster (1990, 1991), Schultze-Berndt & Himmelmann (2004), Himmelmann & SchultzeBerndt (eds.) (2005). Schroeder & al. (2008) provides particularly rich material about functive phrases. However, in most of the literature on secondary predicates, adjectives and non-finite verb forms figure much more prominently than noun phrases, functive noun phrases are not always distinguished clearly from noun phrases in other predicative roles, and even articles that devote a section to the depictive use of nouns do not always provide illustrations of functive phrases. Given that functive-similative syncretism (examined here in Section 4.1) is particularly common, functive phrases have also been discussed in studies mainly 6.  See Schönig (2008) and Schroeder (2008) for a detailed discussion of this construction in Turkish and other Turkic languages.

184 Denis Creissels

devoted to similative constructions — see in particular Haspelmath & Buchholz (1998). By contrast, in the typologically-oriented discussions about the semantic roles fulfilled by noun phrases or adposition phrases in the construction of verbs and their relationships, the functive has been largely ignored so far, with however the notable exception of Hagège (2010: 320–325). The typological literature on semantic roles focuses on participative and circumstantial roles, neglecting the possibility that noun phrases or adposition phrases directly linked to the verbal head of a clause fulfill predicative functions, and that connections exist between predicative roles and participative or circumstantial roles. In this chapter, the functive is mainly investigated as one of the possible roles noun phrases or adposition phrases may fulfill in the construction of verbs, which implies a special emphasis on the possible connections with other roles that can be fulfilled by noun or adposition phrases. The question of the status of the functive in a typology of secondary predicates cannot be ignored, but this perspective is not central to the discussion. For example, the Russian Instrumental is used to encode participants with the semantic role of instrument — Ex. (14a) — but also has predicative uses, in which it is used to mark both nouns in functive role, as in (14b), and adjectives in secondary predicate function, as in (14c). This is certainly not fortuitous, but a proper understanding of this coincidence necessitates also considering the use of the instrumental illustrated in (14d). (14) Russian (p.k.) a. On breet-sja èlektričesk-oj britv-oj. 3sg:m shave:prs:3sg-refl electric-sg.f.instr razor-sg.instr ‘He shaves with an electric razor.’ b. On rabotaet inžener-om. 3sg:m work:prs:3sg engineer-sg.instr ‘He is working as an engineer.’ c. On prišel domoj p’ jan-ym. 3sg:m come:pfv:pst:3sg home drunk-sg.m.instr ‘He came home drunk.’ d. Vse sčitajut ego durak-om. all:pl count:prs:3pl 3sg:m:acc idiot-sg.instr ‘Everybody considers him an idiot.’

According to the analytical grid used in this chapter, the Russian Instrumental has among its possible uses a functive use illustrated by Ex. (14b), and uses in which it expresses other types of identification, illustrated by Ex. (14c–d). In the perspective of this study, this is consequently an instance of a marker whose equative use



Functive phrases in typological and diachronic perspective 185

is not limited to the expression of the functive. The fact that this marker may be found with both nouns and adjectives in the expression of other varieties of identification is just a particular aspect of this polysemy. 2.7 The functive as a comparative concept, and questions of terminology It must be clear from what precedes that the label functive is used here for a comparative concept in the sense of Haspelmath 2010. This implies that no straightforward correspondence must be expected between this notion and the markers used to encode it in individual languages. On the contrary, as illustrated by Ex. (1) to (3) at the beginning of the paper, the general rule is that the markers found with functive phrases also have other uses, sometimes difficult to distinguish from the functive but sometimes very different, and the cross-linguistic variation in the polysemy patterns in which functive markers are involved is amazing at first sight. The only justification for the introduction of the notion of functive is that it may help to discover regularities in the polysemy patterns in which the expression of the functive is involved and to analyze them. Relatively few descriptive traditions acknowledge the importance of the notion of functive in the organization of case or adposition systems. The main exception is the Uralistic tradition. Most descriptions of Uralic case systems mention an Essive case whose uses include the expression of functional identification.7 The functive use of the Essive case of Finnish has already been illustrated in (5). Ex. (15) and (16) illustrate the functive use of the Essive case of Hungarian and Estonian. (15)

Hungarian (Lelkes 1967: 558) Öt év-ig mérnök-ként  dolgoz-ott egy gyár-ban. five year-term engineer-ess work-pst[3sg] one factory-iness ‘He worked five years as an engineer  in a factory.’

(16)

Estonian (Lutkat & Hasselblatt 1993: 192) Minu mees töötab arsti-na . 3sg:gen husband work:prs:3sg doctor-ess ‘My husband works as a doctor .

The fact that Uralic grammars present the functive use as central to the meaning of Essive cases must however not lead one to forget that the expression of functional identification (and possibly other varieties of identification) is not the only use of Uralic Essives, and that the additional uses they may have vary across Uralic languages. For example, the Essive case of Finnish also marks some time adjuncts, 7.  For general orientation about Uralic Essives, see de Groot (2010).

186 Denis Creissels

whereas the Essive case of Hungarian marked by the suffix -ként is also marginally found in similative function, as in sas-ként repül ‘fly like an eagle’ (Balogh 2000: 199). Moreover, even within the Uralic family, the cases regularly used to express the functive and other meanings typically expressed by Uralic Essives are sometimes labeled differently. For example, in Mordvin, the functive is expressed by the so-called Translative case. It may also happen that, in languages in which the Essive case constitutes the most usual expression of the functive, like Hungarian, other cases have functive uses. Ex. (17) illustrates the functive use of the Hungarian Dative. (17)

Hungarian (Anna Sörés, p.c.) Kábítószer-nek  használják ez-t a növény-t. drug-dat use:prs:3pl:def dem-acc def plant-acc ‘This plant is used as a drug .’

Outside the Uralic family, essive is mainly used for cases expressing location, and there is even more variety in the labeling of cases that have the expression of the functive among their possible uses. None of the labels found in descriptive grammars can be viewed as indicative of the existence of a functive use. In Russian, the expression of the functive is one of the possible functions of the Instrumental (see Ex. (14) above), and in Kamaiurá (Tupi-Guarani), it is one of the possible functions of a suffix labeled attributive case (caso atributivo) by Seki (2000). Basque is probably the most extreme case, since in most Basque grammars, a case specialized in the expression of functive and other types of identification is misleadingly labeled prolative. 2.8 The morphosyntactic nature of functive markers Functive phrases are not always marked by elements straightforwardly identifiable as case markers or adpositions. Functive phrases are often marked by multifunctional grammatical words operating not only at noun phrase level, but also at clause level. This is in particular the case of English as. Purépecha — Ex. (18) — provides another example of this kind of situation, with a grammatical word used as a functive preposition, as in (18a), but also as a complementizer, as in (18b). (18) Purépecha (Claudine Chamorreau, p.c.) a. Ari-sïn-ka=kini eski chi amiguti. say-hab-asrt1/2=2sg eski 2sg friend ‘I am talking to you as your friend . b. Wiri, juchiti tapitxu-ni ari eski ju-a-ka ! run[imp] 1sg uncle-acc say[imp] eski come-fut-subj ‘Run and tell my uncle that he should come !’



Functive phrases in typological and diachronic perspective 187

There is no homogeneity in the categorization and labeling of such multifunctional grammatical words by grammarians. For example, in the French tradition, comme ‘as, like’ is labeled ‘conjunction’ even in uses in which this word introduces a noun phrase which can hardly be analyzed as the reduced form of a clause, whereas in similar situations, the labels ‘relative adverb’ or ‘preposition’ can be found in grammars and dictionaries of other European languages. Another difficulty in analyzing the morphosyntactic nature of functive markers follows from the fact that functive noun phrases are not referential, and consequently are commonly devoid of the determiners whose presence characterizes fully-fledged noun phrases. Functive noun phrases are often reduced to bare nouns, even in languages in which the general rule is that every noun phrase must include an overt determiner. For example, in French — Ex. (19), noun phrases introduced by comme ‘as, like’ follow the general rule (obligatory presence of an overt determiner) when comme marks event-oriented similative adjuncts, but do not follow the rule when comme is used as a functive marker, and this constitutes precisely the formal clue to the choice between the functive and similative readings of comme in French. (19) French (p.k.) a. Il parle comme un médecin. 3sgm speak:prs:3sg like indef:sgm doctor ‘He speaks like a doctor.’ (similative: the way he speaks is similar to the way doctors speak) b. Il parle comme médecin. 3sgm speak:prs:3sg like doctor ‘He speaks as a doctor.’ (functive: what he says is in conformity with the role of a doctor)

The predicative nature of functive phrases tends therefore to blur the distinction between case markers or adpositions (combining with fully fledged noun phrases) and derivative morphemes (combining with bare lexemes), and this may be the source of discrepancies between grammarians as to the morphosyntactic nature of functive markers. For example, É. Kiss & al. (2003: 202) express reservations about identifying the essive suffixes of Hungarian -ként and -ul/ül as cases. Examples such as barátjá-ul [friend-poss[3sg]-ess] ‘as his/her friend’ or a gyár főmérnök-e-ként [def factory main_engineer-poss-ess] ‘as the main engineer of the factory’ show however that these suffixes are compatible with modifications typical of nouns heading noun phrases, and consequently must be analyzed as case suffixes rather than suffixes deriving adverbs from nouns.

188 Denis Creissels

2.9 Functive phrases and agreement One consequence of the predicative nature of functive phrases is that they generally cannot show number variation independently of the number of their controller. They either do not show number variation at all, or follow more or less strict number agreement rules. The documentation I have gathered includes no instance of case agreement of a functive phrase with its controller. According to Schultze-Berndt & Himmelmann (2004), case agreement of noun phrases in depictive function is common in Australian languages, but all the examples they give involve varieties of depictives other than functive. 2.10  Zero-marked functive phrases Ex. (20) to (23) can be analyzed as including functive noun phrases devoid of any mark of their role in the clause. (20)

Hindi (Annie Montaut, p.c.) Usne mujhe sahâyak cunâ 3sg:erg 1sg:dat collaborator choose:pst ‘He chose me as (his) collaborator.’

(21)

Arhavi Laz (René Lacroix, p.c.) Ma xezmek’yari kemzdi! 1sg servant take:imp:1sg ‘Take me as your servant!’

(22)

Chechen (Zarina Molochieva, p.c.) Iza imaam xaerzhina. 3sg imam choose:prf ‘They chose him as (their) imam.’

(23) Basque (a: Hualde & Ortiz de Urbina 2003: 446, b: Céline Mounole, p.c.) a. Jon artzain joan da. Jon shepherd go:incmp prs.3sg ‘Jon went (to work) as a shepherd.’ b. Artzainak zakurra izaten du lagun bakarra. shepherd:sg:erg dog:sg have:icmp prs:3sg:3sg companion only:sg ‘The shepherd has a dog (as his) only companion.’

Interestingly, all the examples of zero-marked functive phrases I have been able to gather illustrate situations in which the distinction between functive and individual-level identification tends to blur, since the presumed functive phrases in



Functive phrases in typological and diachronic perspective 189

the examples in question are headed by nouns referring to professions or relatively stable interpersonal relationships. Moreover, in the languages in which I came across zero-marked functive phrases, they are possible with very limited sets of verbs. For example, in the documentation I have been able to gather, none of the translational equivalents of I am talking to you as your friend includes a zeromarked noun phrase. This can be viewed as evidence that such functive phrases form part of the argument structure of the verbs with which they are found. 2.11 The syntactic role of the controller of functive phrases Himmelmann & Schultze-Berndt (2005: 54–55) discuss the question of the restrictions on the syntactic role of the controller of participant-oriented adjuncts. After noting that “for depictive secondary predicates, it is generally assumed that they can be controlled by subjects or objects or, more precisely, by the single core argument of an intransitive predicate (S), the actor-like core argument (A) of a transitive predicate, and the undergoer-like argument (O) of a transitive predicate,” they observe that “languages differ considerably with regard to which participants are easily accessible to controllers,” and that for a single language “there appear to be considerable differences with regard to acceptability and usage depending on speaker, context, genre, and medium (spoken vs. written usage)”. In the particular case of functive phrases, the documentation I have gathered confirms that functive phrases are commonly controlled by S, A, or O, and that no clear ergative-like or accusative-like patterning (i.e., exclusion of either A or O from the set of the possible controllers of a functive phrase) emerges in the languages for which I have sufficient data. I am not in a position to put forward any generalizations concerning the control of functive phrases by terms other than A, O, or S. 3. Functive, equative, and transformative 3.1 Introductory remarks Independent equative clauses may or may not include a copula. Noun phrases in predicate function in independent equative clauses are commonly devoid of any mark of their syntactic role, but in some languages they occur in a marked case form or combined with one of the adpositions used to introduce oblique terms in the construction of verbs. It is relatively common that the expression of dependent equative predications brings into play special secondary predicate constructions rather than the

190 Denis Creissels

constructions available for the expression of main predications. As illustrated by Ex. (24), in French, depending on the semantic relationship with the main predication, noun phrases in secondary predicate role may not require the use of the copulative verb être, but the presence of a preposition may be necessary. (24) French (p.k.) a. Enfant  il était souvent malade. child he was often ill ‘He was often ill as a child.’ (= when he was a child) b. En_tant_que Français  il a du mal avec l’accentuation inasmuch_as Frenchman he has some difficulty with the-stress de l’anglais of the-English ‘As a Frenchman  he has difficulties with English stress.’ (= since he is a Frenchman) c. Il parle bien anglais pour un Français . he speaks well English for a Frenchman ‘He speaks English well for a Frenchman .’ (= although he is a Frenchman)

3.2 Markers shared by functive phrases and noun phrases in secondary predicate function expressing other types of identification It is common that markers used in typical functive situations also occur with noun phrases involved in secondary predications expressing other types of identification. This is in particular the case for English as in its use as a preposition. Depending on the context, as an N, in addition to its functive reading (‘in the role of an N’), may also be interpreted as ‘when (s)he was an N’ or ‘since (s)he is an N’. Ex. (25) to (27) provide additional illustrations of the use of the same case forms or adpositions (the Estonian Essive, the Russian Instrumental, and the Spanish preposition de) with functive phrases and noun phrases expressing other types of identification, in particular with a meaning of temporally bound identification. (25) Estonian (Lutkat & Hasselblatt 1993: 192) a. Minu mees töötab arsti-na. 3sg:gen husband work:prs:3sg doctor-ess ‘My husband works as a doctor.’ b. Poisi-na mängisin jalgpalli. boy-ess play:pst:1sg soccer:part ‘As a boy I played soccer.’



Functive phrases in typological and diachronic perspective 191

(26) Russian (p.k.) a. Ja pomnju ego učitel-em. 1sg remember:prs:1sg 3sg:sg:m:acc teacher-instr ‘I remember him as a teacher.’ b. Mal’čik-om ja igral v futbol. boy-instr 1sg play:pst:sg:m in soccer:acc ‘As a boy I played soccer.’ (27) Spanish (p.k.) a. De postre comimos toronjas. de dessert eat:pst:1pl grapefruit:pl ‘We ate grapefruit as dessert.’ b. Juega de centro delantero. play:prs:3sg de center forward ‘He plays (as a) center forward.’ c. De niño era muy rebelde. de child be:pst:3sg very rebellious ‘As a child (= when he was a child) he was very rebellious.’ d. Yo de ti no lo haría. 1sg de 1sg neg 3sg:m do:cond:1sg ‘If I were you I would not do that.’ lit. ‘I as you would not do that.’

3.3 Functive markers and predicate marking in independent equative clauses In independent clauses in which the main predicate is a noun phrase expressing the identification of an entity (This is a tree, This tree is an oak), languages may or may not make use of copulas, but in both cases, the use of a marked case form or adposition with noun phrases in predicate function in independent equative clauses is relatively uncommon in the languages of the world. There are however languages in which nouns expressing equative predication in independent clauses may occur in the same marked case form or combined with the same adposition as in functive role — Ex. (28). (28) Avar (Gimbatov 2006) a. Wakil-ɬun w-it’-ana. delegate-ɬun m-send-cmp ‘He was sent as a delegate.’ b. Coħo mast’i b-uk’-ana hesije halmaʁ-ɬun. only dog n-be-cmp he:dat companion-ɬun ‘A dog was his only companion.’ (lit. ‘Only a dog was to him as a companion.’)

192 Denis Creissels

In most cases, the use of nouns in predicate function in independent equative clauses with the same marking as in functive secondary predicate function competes with the use of zero-marked noun phrases. This is the case in Avar, but also in Finnish (see Ex. (5) above), Russian, etc. In such cases, the use of an overt marker with noun phrases in predicate function in independent equative clauses tends to correlate with non-permanent (or stage-level) identification, whereas zero marking tends to be preferred with reference to permanent (or individual-level) identification — see in particular Erelt & Metslang (2003) on Estonian and Turunen (2011) on Mordvin. A different pattern is however found in Mandinka and other Mande languages, in which independent equative clauses include a postposition also used as a functive marker in conditions that are entirely determined by syntax and have absolutely no semantic correlate. The use of the Mandinka postposition tí as a functive marker and in independent equative clauses has already been illustrated in Ex. (3a) & (3f), reproduced here as (29a–b). Ex. (29c) illustrates independent equative clauses without tí. The rule is that tí is obligatory if the entity whose identification is expressed is explicitly designated by means of a noun phrase (ñiŋ yírôo ‘this tree’ in (29b)), whereas it does not occur in identification clauses with no indication of the entity to which an identification is attributed, as in (29c). This example shows that semantics plays no role in this choice: tí is obligatory in identification clauses expressing permanent identification too, provided the entity whose identification is expressed is explicitly designated. (29) Mandinka (p.d.) a. Ŋ́ ŋá até le lóŋ ŋ́ faamáa ti. 1sg cmp.pos 3sg:emph foc know 1sg father:def obl ‘It is him that I knew as my father.’ b. Ñiŋ yíróo mu bantáŋo le ti. dem tree:def cop silk_cotton_tree:def foc obl ‘This tree is a silk-cotton tree.’ c. Bantáŋo le mu. silk-cotton_tree:def foc cop ‘This is a silk-cotton tree.’

3.4 Functive-transformative syncretism The use of the Mandinka postposition tí not only with functive phrases and in independent equative clauses but also as a transformative marker has already been illustrated in Ex. (3). Similarly, the Avar suffix -ɬun illustrated in Ex. (26) above



Functive phrases in typological and diachronic perspective 193

in functive marker function and in independent equative clauses is also used as a transformative marker — Ex. (30). (30)

Avar (Gimbatov 2006) Dow rosdal administracijaɬul bet’er-ɬun k̄un wugo. dem:m village:gen administration:gen head-ɬun fall:cvb m:cop ‘He became (lit. ‘fell’) the head of the village administration.’

Mordvin (Uralic) provides an additional illustration of functive-transformative syncretism. In Finnish, functive and transformative are expressed by two distinct cases, the Essive and the Translative, respectively. In Mordvin, the so-called Translative case, marked by a suffix -ks cognate with the Translative case of Finnish, is found both in transformative and functive marker function — Ex. (31).8 (31) Erzya Mordvin (Grünthal 2003: 185) a. Son roboti vračo-ks. (s)he work:3sg doctor-transl ‘(S)he works as a doctor.’ b. Ekšeś vel’avtś kel’me-ks. weather:def change:refl.pass:impf.3sg cold-transl ‘The weather became cold.’

A similar situation is found in the Tupi-Guarani family too, with a suffix labeled Attributive in Seki (2000) but more commonly labeled Translative in descriptions of Tupi-Guarani languages. This suffix is reconstructed as *-(r)amõ in Proto-TupiGuarani (Françoise Rose, p.c.). For descriptions of its use in Tupi-Guarani languages, see among others Rose (2011: 235–240), Seki (2000: 110–112). 4. Functive and other non-participative roles 4.1 Functive-similative syncretism The various types of grammatical elements used as functive markers may have another use, termed similative here, in which their combination with a noun phrase N is interpreted as ‘like an N’, ‘in the same way as an N’. Functive-similative syncretism has been systematically investigated by Haspelmath & Buchholz (1998) for the languages of Europe, and is so common, not only in the languages of Europe, but also at the global level, that one might wonder whether functive and similative must be defined as distinct roles rather than being considered as two particular 8.  See Riese (1992–1993) for on overview of Translative in Uralic.

194 Denis Creissels

cases of a more general notion. However, it is easy to find languages in which prototypical functive and prototypical similative are not encoded in the same way. This is the case in English with as and like, and a similar distinction is found among others in Greek — Ex. (32). (32) Greek (Holton & al. 1997) a. Sou miláo os fílos kai óchi os giatrós. 2sg:gen speak:prs:1sg as friend and neg as doctor ‘I am speaking to you as a friend and not as a doctor.’ b. Μiláei san dikigóros. speak:prs:3sg like lawyer ‘He speaks like a lawyer.’

The justification for distinguishing functive and similative relies on the general semantic distinction between secondary predicates and verb modifiers: noun phrases in functive role express a predication about a participant involved in the event denoted by the verb, whereas noun phrases in similative role can usually be analyzed as a particular type of manner adjunct. However, pragmatically, the distinction is not so clear-cut, since ‘x does something in the way N’s do’ does not imply ‘x fulfills the role of an N’, but does not exclude it either, which leaves the door open to the reanalysis of ‘do something like an N’ as ‘do something as an N’. We will see in Section 8 that this process is attested or can be reconstructed in the history of many languages. Moreover, a notion that can be labeled participant-oriented similative, paraphrasable as ‘having taken on the appearance of ’, can be recognized in sentences such as She appeared on stage as a man. In this sentence, as a man is clearly a participant-oriented adjunct, which does not specify a particular way of appearing on stage but makes explicit the appearance taken on by the woman in question at the moment she appeared on stage. What is relevant here is the contrast with plain (or event-oriented) similative, illustrated by She appeared on stage like a man, which describes a situation in which a woman is clearly recognizable as a woman but does something in a way considered typical for men. Participant-oriented similative has in common with plain (or event-oriented) similative the notion of similarity, whereas it has in common with functive the orientation towards a participant. It is particularly interesting that English, which uses distinct markers for plain similative (like) and functive (as) does not mark participant-oriented similative with like, but with as. More generally, it is easy to imagine transitional situations that may facilitate the extension of a similative marker to functive marking, or vice-versa, which explains why functive-similative syncretism is so widespread.



Functive phrases in typological and diachronic perspective 195

Note however that the use of the same marker with functive and similative phrases does not necessarily imply ambiguity at the construction level. In the Zulu example (33), the construction gives no formal clue to decide whether njengamust be interpreted as the equivalent of English as or like. By contrast, in the Spanish Ex. (34), the same marker como can be used in both functions, but the distinction manifests itself in the use of determiners (no determiner in (34a), where como introduces a functive noun phrase vs. indefinite article in (34b), where como introduces a similative noun phrase). A situation similar to that of Spanish has already been illustrated above for French — see Section 2.8, Ex. (17). (33) Zulu (Online Zulu-English dictionary, http://isizulu.net) a. Ngi-sebenza njeng-omshayeli wa-matekisi.9 1sg-work as/like-cl1:driver cl1:gen-cl6:taxi ‘I work as a taxi driver.’ b. U-khala njeng-omfazi.10 2sg-speak as/like-cl1:woman ‘You speak like a woman.’ (34) Spanish (p.k.) a. Te lo digo como médico. 2sg.dat 3sgm.acc tell:pres:1sg as/like doctor ‘I am telling that to you as a doctor.’ b. Hablas como un médico. speak:prs:2sg as/like indef doctor ‘You are speaking like a doctor.’

In addition to languages having a single marker equally productive in functive and similative uses (like Spanish como or Zulu njenga), many languages have productive functive markers marginally found in similative uses, or productive similative markers marginally found in functive uses. For example, in Russian, the Instrumental case, whose functive use illustrated in (14b) above is relatively productive, also has a more marginal similative use (as in vyt’ volkom ‘howl like a wolf ’). Functive-similative syncretism often involves markers that cannot be straightforwardly identified as cases or adpositions and probably developed from similative subordination, since they also have uses in which they operate at clause level (see Section 8). In this connection, it may be interesting to compare the use of such multifunctional markers in adverbial subordination with their use as functive markers. 9.  Vowel merging (a + u > o) occurs at the boundary between njenga- and umshayeli ‘driver’. 10.  Vowel merging (a + u > o) occurs at the boundary between njenga- and umfazi ‘woman’.

196 Denis Creissels

For example, in Spanish, both Estoy aquí de  médico and Estoy aquí como  médico are possible equivalents of English I am here as a doctor. However, according to Spanish dictionaries, the construction with como ‘like, as’ followed by a bare noun also has an explicative identification reading (‘I am here because I am a doctor’), whereas this reading is not available with de, and it is tempting to hypothesize a relationship with the fact that como as a conjunction can be used to introduce explicative clauses (as in como eres médico ‘since you are a doctor’). Interestingly, de Groot (2008) suggests a similar distinction in Hungarian between the Essive case and the use of mint ‘like’. 4.2 Functive and comparative of superiority Contrary to the languages mentioned in Section 4.1, Mandinka uses different markers in functive and similative functions, but in the Mandinka construction expressing the comparative of superiority, the standard of comparison is marked by the postposition tí, also used as the marker of functive and other non-participative roles, as already illustrated in Section 2, Ex. (4a). The same syncretism is found among others in German — Ex. (35) — and Latvian — Ex. (36). (35) German (Ekkehard König, p.c.) a. Das sollte als Beispiel dienen. that should as example serve ‘That should serve as an example.’ b. Gold ist schwerer als Silber. gold is heavier than silver ‘Gold is heavier than silver.’ (36) Latvian (Veksler & Jurik 1975: 86, 94) a. Viņš strādā par skolotāju. he works as teacher ‘He works as a teacher.’ b. Zęlc ir smagāks par sudrabu. gold is heavier than silver ‘Gold is heavier than silver.’

4.3 Functive and location The Essive case of Finnish, in addition to its particularly productive use as a functive marker, is used with nouns that denote time to express temporal location, and is also found in a few lexicalized expressions of spatial location (which originally

Functive phrases in typological and diachronic perspective 197



constituted the central and most productive function of this case), for example kotona ‘at home’, kaukana ‘afar’.11 The French preposition en ‘in’ (< Latin in) has lost much of its productivity as a spatial marker, since in Modern French, interiority is more commonly expressed by means of dans < de intus ‘from within’, but en has developed many non spatial uses. In particular, en marks the argument of verbs such as consister en ‘take the form of ’, se comporter en ‘behave as’, traiter en ‘treat as’, changer en ‘change into’. Although less commonly found than comme in this function, en is also used to introduce functive adjuncts — Ex. (37). (37)

French (p.k.) Je vous parle en ami. I to.you speak in friend ‘I’m speaking to you as a friend.’

4.4 Functive and manner In Kartvelian languages (with the exception of Laz), the same suffix (traditionally labeled Adverbial case) is found with noun phrases in functive role (38a), in transformative role (38b), or expressing embedded equative predications (38c), and in the derivation of manner adverbs from adjectives (38d). According to Boeder (2005), most of the manner adverbs formed by means of the Adverbial case “seem to have a semantic subject orientation”. (38) Georgian (Boeder (2005) and Manana Topadze, p.c.) a. Masc’avlebl-ad mušaobs. teacher-adv work:prs:3sg ‘He/she works as a teacher.’ gadaikca. b. ɣvino ʒmr-ad wine vinegar-adv change:aor:3sg ‘The wine changed into vinegar.’ c. K’arg ekim-ad itvleba. good doctor-adv be_counted:prs:3sg ‘He/she is considered a good doctor.’ d. Mat’arebeli zant’-ad daiʒra. train lazy-adv moved ‘The train moved lazily.’

11.  For a detailed analysis of the uses of the Finnish Essive, see Onikki-Rantajääskö (2006).

198 Denis Creissels

Suffixes with the same distribution are also found in North West Caucasian languages — Ex. (39).12 (39) Adyghe (Arkad’ev & al. 2009: 55) a. Sjezdəm djeljegat-ew wəqeḳʷaʁeš’tən. to_the_meeting delegate-adv you_probably_came ‘You probably came to the meeting as a delegate.’ b. Aš’ əš-ew se səχʷəʁ. his brother-adv I I_became ‘I became his brother.’ c. Dax-ew matxe. beautiful-adv he_writes ‘He writes beautifully.’

Note however that, at least in Georgian, the functive use of the Adverbial case is subject to restrictions about which I am unfortunately not able to be more precise. For example, in English, as is equally used in work as a doctor and say something as a doctor, but according to my Georgian consultant, in Georgian, the Adverbial case can be used only with ‘work’, and rogorc ‘like, as’ must be used with ‘say’ — Ex. (40). (40) Georgian (Manana Topadze, p.c.) a. Ekim-ad  mušaobs. doctor-adv work:prs:3sg ‘He/she works as a doctor.’ b. Me amas geubnebi šen ara rogorc ekimi , 1sg this:dat say:prs:1sg:2sg 2sg neg as doctor ‘I am saying this to you not as a doctor aramed rogorc megobari. but as friend but as a friend.’

Dargi (North East Caucasian) has a suffix -li which, in addition to the uses typical of South and North West Caucasian Adverbial cases, also marks a converbial form of verbs — compare qaraul-li ‘as a guardian’, wana-li ‘warmly’, uku-li ‘eating’ (Musaev 1987: 70–72). Among Lezgic languages (another branch or North East Caucasian), Archi has a suffix -ši with a similar distribution. This suffix productively marks the functivetransformative form of nouns (as in dijark̄a-ši ‘as a milkmaid’) and a converbial form of verbs (as in lap’a-r-ši ‘throwing’, where -r- is an imperfective marker), and is 12.  See Vydrin (2008) for a detailed presentation of the uses of the suffix -ew in Adyghe.



Functive phrases in typological and diachronic perspective 199

also found in some adverbs derived from nouns or adjectives (as in kutak-ši ‘strongly < kutak ‘strength’). But in addition to that, -ši also takes part in another type of syncretism (the functive-lative syncretism — see Section 5.3), since in the system of spatial cases, -ši encodes destination of movement (as in perma-l-a-ši ‘into the farm’, where -l- is the oblique stem marker, and -a- the localization marker ‘in’). 5. Functive and participative roles 5.1 Functive, instrumental, and comitative The Instrumental case of Russian has not only an instrumental use, but also predicative uses, including functive, as already illustrated by Ex. (14) above. A similar situation is found in Lithuanian, with a case labeled Instrumental, which has instrumental (41a) and perlative (41b) uses but which is also widely used to mark functive and other predicative uses of nouns (41c–f). (41) Lithuanian (a-b: Hewson & Bubenik 2006: 209, c-f: Nichols 1983) a. Jìs rãšo pieštukù. he writes pencil:instr ‘He writes with a pencil.’ b. Jìs cĩna keliù. he walks road:instr ‘He walks along the road.’ c. Jis dirba mokytoju. he works teacher:instr ‘He works as a teacher.’ d. Karas padarė jį luošu. war made him invalid:instr ‘The war made him an invalid.’ e. Jis grįžo didvyriu. he returned hero:instr ‘He returned a hero.’ f. Aš jį laikau kvailiu. I him consider fool:instr ‘I consider him a fool.’

Functive-instrumental syncretism is also found in the West Mande language Soso. The Soso postposition rá is used in the same way as the Mandinka postposition tí (see above) in functive-transformative function and in identificational predication, and is also productively used in instrumental function. In contrast with Mandinka tí, Soso rá is not used in the construction expressing the comparative of

200 Denis Creissels

superiority (in this function, Soso uses the benefactive postposition bɛ́). Rá is also marginally found in spatial and comitative uses. The Mandinka postposition tí is mainly used to encode various non-participative roles (including functive), and is not used at all in instrumental function, but it is marginally found with a comitative meaning, as already illustrated by Ex (4b), Section 2.1. Among West Mande languages, a postposition ni that is probably cognate with Mandinka tí, but with a wider range of uses, is found in Bozo. Judging from the examples in Daget et al. (1953), in addition to its functive-transformative use and to its use in equative predication, ni has at least some productivity in instrumental, comitative, and spatial uses. 5.2 Functive, benefactive, and substitutive Ex. (42) illustrates the use of the Bambara postposition yé, fully productive in benefactive function (42a) and also used as a functive marker (42b) and in constructions that can be analyzed as involving an embedded equative predication (42c). This postposition is also used to encode the addressee of speech verbs (42d) and the recipient of some transfer verbs, but not purposive adjuncts, for which another postposition (káma) is used. (42) Bambara (p.d. & Gérard Dumestre, p.c.) a. Ù yé báara kɛ́ ń yé. 3pl cmp.pos work do 1sg ye ‘They worked for me.’ b. Ù yé à cí làsigiden yé. 3pl cmp.pos 3sg send ambassador ye ‘They sent him as an ambassador.’ c. Ń bɛ́ à jàti nálonma yé. 1sg incmp.pos 3sg consider idiot ye ‘I consider him an idiot.’ d. Ù yé tìnyɛ fɔ́ í yé. 3pl cmp.pos truth tell 2sg ye ‘They have told you the truth.’

Among West Mande languages, a similar situation is found in Bobo, with a postposition nā used in functive-transformative function but particularly productive in benefactive function (Le Bris & Prost 1981: 75). Benefactive-functive syncretism may at first sight seem puzzling, since the closeness between these two notions is not obvious. However, benefactive markers often have a substitutive use (sometimes also called surrogative). In English,



Functive phrases in typological and diachronic perspective 201

do something for N can be interpreted as do something in place of N, depending on the context, and it is not difficult to conceive of the possibility of a shift from do something in place of N to do something in the role of N, since the difference is essentially a question of referentiality. The development of a functive use of the preposition za introducing an Accusative noun phrase in Lower Sorbian, described by Bartels (2008), supports the hypothesis of a direct link between substitutive and functive, since among Slavic languages, substitutive is the most common non-spatial use of za (whose original meaning is ‘behind’) combined with the Accusative case. The Latvian preposition par, whose use as a functive marker and in the expression of the comparative of superiority has been illustrated in Ex. (36) above, must also be mentioned here, since it is not used as a benefactive marker but is found in contexts in which it can be analyzed as expressing a particular variety of substitutive — Ex. (43). (43) Latvian (Veksler & Jurik 1975: 94) a. Es jegādājos grāmatu par rubli. I got book for ruble ‘I got the book for one ruble.’ b. Viņš maksās par visu. he will_pay for everything ‘He will pay for everything.’

In Mandinka too, the postposition tí, used as a functive marker, has no benefactive use, but is found marginally with this particular variety of substitutive meaning, as already illustrated by Ex. (4c), Section 2.1. Interesting data about functive-benefactive syncretism can be found in Basque as well. The Benefactive case of Basque (sometimes called Destinative) is used exclusively to encode beneficiaries; its uses do not overlap with those of the Dative and Causal (or Motivative) cases, and it shows no formal similarity with them either. By contrast, the relationship between the Benefactive case and the misnamed Prolative case (a case used exclusively with nouns in secondary predicate function) is obvious, since the ending of the Benefactive -(r)entzat can be decomposed as -(r) en (Genitive) + -tzat (Prolative). Moreover, as illustrated by Ex. (44), the distinction is partially neutralized for personal pronouns, which can occur in the Benefactive form with meanings that normally require the use of the Prolative form. (44) Basque (Lafitte 1962: 439) a. Zu-retzat egin dut. 2sg-ben do:cmp prs:3sg:1sg ‘I did it for you (in your interest, or instead of you).’

202 Denis Creissels

b. Zu-retzat utzi dut. 2sg-ben leave:cmp prs:3sg:1sg ‘I confused him/her with you.’ lit. ‘I left him/her as you.’

Given that, in the noun inflection of Basque, some case endings include an optional formative identical to the Genitive suffix (as in Edurne(ren)gan, locative of the proper name Edurne),13 it seems reasonable to view this formal similarity and possibility of neutralization as evidence that the Basque cases used to mark benefactive and predicative noun phrases result from the specialization of two suffixes that originally were variants of the same case or postposition. 5.3 Functive and lative (destination of movement) In some Uralic languages (Mari, Southern Khanty), cases described as primarily encoding the destination of movement (lative) are also used in predicative functions, including functive. However, the cases in question also occur in the expression of transformation and location, and consequently do not provide evidence for a direct link between functive and the expression of destination of movement, since lative-locative, lative-transformative, and transformative-functive syncretism are all very common cross-linguistically. Interestingly, as mentioned above (Section 4.4), Archi (East Caucasian) has a suffix -ši, marking the functive-transformative and lative forms of nouns, but also found as the mark of a converbial form of verbs and in the derivation of manner adverbs. 5.4 Functive and separative (source of movement) In Spanish, some predicative uses of nouns, including functive, can be marked by the preposition de — Ex. (27) above and Ex. (45). This preposition has a wide range of possible uses, but it developed from the Latin preposition de ‘down from’, and is still fully productive with its original meaning. (45) Spanish (p.k.) a. Juega de centro delantero. play:prs:3sg de center forward ‘He plays (as a) center forward.’ 13.  A possible explanation of this phenomenon is that the Basque cases including an optional formative identical to the Genitive suffix were originally postpositions governing the Genitive and that either the Genitive marking was not obligatory or the deletion of the Genitive suffix became possible during the process by which postpositions were converted into case suffixes.

Functive phrases in typological and diachronic perspective 203



b. c.

De niño era muy rebelde. de child be:pst:3sg very rebellious ‘As a child he was very rebellious.’ Yo de ti no lo haría. 1sg de 1sg neg 3sg:m do:cond:1sg ‘If I were you (lit. ‘as you’) I would not do that.’

Given the wide variety of the possible uses of de (which among others has instrumental uses, although the instrumental is more commonly expressed by means of con ‘with’), it is difficult to decide whether a direct link must be posited or not between its original separative meaning and its predicative uses, and I have not been able to find studies dealing with the historical aspect of this question. In particular, it is not clear to what extent the development of the uses of de in early Romance may have been influenced by the polysemy pattern of the Ablative case of Latin, which among others included the expression of instrument, manner, comparison, and cause. The Italian preposition da has uses that coincide only partially with those of Spanish de. In particular, it may be used to encode location or destination, which is not possible for Spanish de. However, Italian da also has the encoding of the source of movement as its primary function, and its uses as a functive marker are quite comparable to those of Spanish de. 5.5 Others Nenets has a case labeled Essive which, in addition to functions commonly fulfilled by Uralic Essives (including the expression of the functive), also has uses whose relationship to the other uses of this case form is unclear. According to Irina Nikolaeva (p.c.), it may mark emphatic subjects, “indicating some kind of contrast between the referent and a potential set of other relevant candidates, sometimes with an additional exclamative meaning (admiration)” — Ex. (46). (46)

Nenets (Irina Nikolaeva, p.c.) Wera-ŋe° əwa=n’uq! Wera-ess good-excl ‘But Wera is so nice!’

The Nenets Essive may also function as an address (vocative), “again when other people are present and there is contrast” (Irina Nikolaeva, p.c.). It is not clear whether this contrastive use of the Nenets Essive must be viewed as an extension of one of the uses typical for Uralic Essives, or is rather a vestige of the origi-

204 Denis Creissels

nal meaning of the element that has grammaticalized as a functive marker in the history of Nenets. 6. Complex situations and their interpretation Several of the examples mentioned in the previous sections raise a problem that considerably complexifies the study of the semantic affinities between functive and other semantic roles: most of the time, the functive markers also used to encode some participant roles have very wide ranges of possible uses, among which it does not seem possible to select a particular role that could safely be considered as having a particularly direct relationship with the functive. An additional illustration is provided by the Korean ‘directional’ case particle (u)lo, which according to Sohn (1999: 337) marks “direction in the sense of ‘towards, to, for’, instrument in the sense of ‘with, of, by, in’, and function in the sense of ‘as, for, in the capacity of, in terms of ’.” A possible clue to this puzzle could be provided by cross-linguistic observations revealing marked differences in the frequency with which other roles occur in polysemy patterns involving the functive. Unfortunately, the polysemy patterns I have been able to observe so far vary in such a way that the mere frequency with which the various syncretisms occur among languages does not permit the selection of a particular participant role as having more affinities with functive than the others. The following observations will, however, guide the proposals formulated in Sections 7, 8, and 9: – Functive markers whose non-functive uses do not go beyond the expression of equative and/or transformative are relatively rare, and all those I came across can be analyzed as resulting from one of the grammaticalization processes presented in Section 7. – Among the polysemy patterns in which functive markers are involved crosslinguistically, there is a relatively clear-cut distinction between polysemy patterns involving similative but neither locative nor participant roles, and polysemy patterns involving locative or participant roles, which suggests that specific grammaticalization patterns are responsible for the emergence of similative-functive markers. – The polysemy patterns involving locative and/or participant roles are found, most of the time, with adpositions or cases with a puzzling variety of uses, whose diachronic development is extremely difficult to reconstruct in the absence of historical data; this can be viewed as evidence that functive uses tend to appear



Functive phrases in typological and diachronic perspective 205

at relatively late stages in the evolution of adpositions or cases, since if this were not the case, less intricate polysemy patterns should be more common. – In all the cases I came across in which the origin of a functive marker with a polysemy pattern involving locative and/or participant roles is known or easily reconstructable, the original meaning is spatial (locative, lative, or separative). 7. Grammaticalization paths directly resulting in the emergence of functive markers 7.1 ‘In the quality of ’ > functive marker In some languages, nouns expressing quality derived from concrete nouns constitute the commonest way to express the meanings expressed in other languages by functive phrases as defined in Section 2.1. For example, in Agul, a language belonging to the Lezgic branch of East Caucasian, nouns derived by means of the abstraction suffix -ʕʷel are widely attested as the Nominative argument of aq’as ‘do’ with the meaning ‘work/act as N’ — Ex. (47a), in the Dative case with the meaning ‘in order to act as N’ — Ex. (47b), and in the Locative case or with the adverbial marker -di with the meaning ‘acting as N’ — Ex. (47c). (47) Agul (Timur Maisak, p.c.) a. K̄andejefij čas müʕelim-ʕʷel aq’as. want:pst we:dat teacher-abstr do:inf ‘We wanted to work as teachers (lit. ‘We wanted to do teacherhood’).’ b. Zun šune dijark̄a-ʕʷeli-s. I go:pfv milkmaid-abstr-dat ‘I went to work as a milkmaid (lit. ‘I went to milkmaidhood.’) c. Čas q̄arawul-ʕʷel-di hataje inebi. we:dat guard-abstr-adv send:ipfv Xinebi ‘They send us Xinebi as a guard (lit. ‘with guardhood’).’

Lezgi has an abstraction suffix -wal cognate with Agul -ʕʷel and lending itself to similar uses. According to Haspelmath (1993: 105), the inessive form of abstract nouns formed by means of this suffix (i.e., a form that can be glossed as ‘in X-hood’) is regularly used to express profession or function, as in reğuxbanwile k’walaxun ‘work as a miller’, lit. ‘work in millerhood’. Moreover, this author discusses evidence suggesting that the complex ending -wil-e [abstr-iness] (where -wil- is an allomorph of the abstraction suffix -wal) is grammaticalizing into a ‘functive’ case suffix.

206 Denis Creissels

7.2 Dependent form of a copulative verb > functive marker Dependent forms of copulative verbs used in constructions of the type illustrated in Turkish by Ex. (13) — reproduced here as (48) — may grammaticalize into more or less opaque secondary predicate markers used in particular with noun phrases in functive role. (48)

Turkish (Göksel & Kerslake: 216) Bun-u san-a avukat ol-arak değil, arkadaş ol-arak söyl-üyor-um. dem-acc 2sg-dat lawyer be-cvb neg friend be-cvb say-prog-1sg ‘I’m saying this to you not as a lawyer but as a friend.’

The languages of the Avar-Andic-Tsezic branch of East Caucasian provide a good illustration of synchronically opaque functive-transformative markers for which a scenario of this type can be reconstructed. For example, in diachronic perspective, the functive-transformative marker of Avar -ɬun (cf. wakil-ɬun ‘as a delegate’ < wakil ‘delegate’ in Ex. (28) above) can be decomposed as -ɬ-un, with a second formative -un cognate with the converbial ending of Avar verbs -un. Synchronically, the segmentations *wakil-ɬ-un or *wakil ɬ-un are ruled out, since no verb *wakil-ɬ-ize or *ɬ-ize is attested. However, in spite of the fact that none of the languages of the Avar-Andic-Tsezic branch of North East Caucasian has a verb whose root could be analyzed as a reflex of *ɬ-, they provide ample evidence that such a root can be reconstructed with the meaning ‘become’, and this reconstruction is widely accepted by specialists — see among others Alekseev (1988: 35). In particular, it is supported by the existence of derivative suffixes such as Avar -ɬ-, used to convert adjectives (and marginally nouns) into verbs. For a more detailed discussion, including a detailed description of the functive-transformative suffix of Northern Akhvakh (Andic), see Creissels (2014). Musaev (1987: 70–72) analyzes the functive-transformative form of nouns in Dargi (a language belonging to another branch of North East Caucasian) as having a similar source, without however providing convincing evidence in support of this analysis. As already mentioned in Section 6 above, the Dargi suffix -li is used both in noun inflection as a functive-transformative marker (as in qaraul-li ‘as a guardian’), and in verb inflection as a converbial suffix (as in uku-li ‘eating’). According to Musaev, a functive form such as qaraul-li ‘as a guardian’ results historically from the reduction of the phrase qaraul wiubli [guardian being], where wiubli is a converbial form of the copulative verb: the stem wiub- was simply deleted, and the suffix -li became attached directly to the noun. However, Musaev does not provide any evidence of a historical process of deletion, and his hypothesis remains purely speculative.



Functive phrases in typological and diachronic perspective 207

The same can be said about the functive-transformative marker -ši of Archi (a language belonging to the Lezgic branch of North East Caucasian). In contrast to Avar-Andic-Tsezic languages, in which there is evidence that the functive-transformative marker of nouns is the reflex of a converbial form of an ancient verb ‘become’, the functive-transformative markers of Dargi and Archi are synchronically transcategorial markers, and there seems to be no clear evidence supporting the reconstruction of a particular historical scenario accounting for their transcategoriality. 8. Grammaticalization paths resulting in the emergence of similativefunctive markers or functive markers with marginal similative uses 8.1 Deictic manner adverbs or words expressing similarity in complex constructions > similative markers > functive markers This grammaticalization path is widely attested among Indo-European languages. Although this is not relevant for the development of such words into functive markers, it can be observed that, in ancient Indo-European languages, words that can be glossed as ‘in this way’ constituted correlative pairs with words glossable as ‘in which way’, as in Latin ut … ita — Ex. (49).14 (49)

Latin (Ernout & Thomas 1953: 355) Ut sementem feceris, ita metes. in_which_way sowing:acc do:subj:prf:2sg in_this_way reap:fut:2sg ‘As you sow, so shall you reap.’

English as comes from Old English alswa ‘quite so’, a compound form whose second formative is the reflex of Proto-Germanic *swae ‘in this way’ (cf. English so, Dutch zo, etc.). ‘In this way’ is also reconstructed as the original meaning of Greek ως ‘as’, which in Ancient Greek had a wide range of uses as a conjunction, but has specialized in Modern Greek as a marker of noun phrases in secondary predicate functions (including functive), distinct from the similative marker σαν ‘like’. In Latin, functive was one of the possible uses of ut ‘in which way’ illustrated in Ex. (49) above. Ut was subsequently replaced by quōmodo, which expressed the same meaning analytically, and whose reflexes (French comme, Spanish como, etc.) are used as similative and functive markers in modern Romance languages. 14.  The glosses ‘in this way’ and ‘in which way’ are given here as pure translational equivalents, and must not be viewed as suggesting that such words necessarily originate from the grammaticalization of expressions involving nouns.

208 Denis Creissels

8.2 ‘In the way / state / shape of ’ > similative marker > functive marker 8.2.1 The case of Maltese bħala Maltese has a functive marker bħala ‘as, in the capacity of ’ illustrated by Ex. (50). (50)

Maltese (Aquilina 1999) Qed inbgħatlek dan irrigal bħala xhieda ta’ mħabbti. prog send:ipfv:1sg:dat2sg prox:sgm present as token gen love:1sgf ‘I am sending you this present as a token of my love.’

According to Martine Vanhove (p.c.), this functive marker derives from bħal, glossed in Aquilina’s dictionary as ‘like, similar to, a sort of, a kind of; as’. Cognates of bħal are used as equivalents of like (and sometimes also as modalizing particles) in Arabic varieties; their common source is the grammaticalization of bi ‘with, by’ + ħāl ‘state’, ‘way’. 8.2.2 The case of Hungarian -ként According to one of the etymologies that have been proposed for the Hungarian Essive suffix -ként (which is mainly used as a functive marker but also has marginal similative uses), this suffix might result from the grammaticalization of an inflected form of kép, a noun borrowed from a Turkic language with the meaning ‘shape’. Interestingly, in Turkic languages, cognates of this noun grammaticalized into similative markers: Turkish gibi ‘like’, Azeri kimin ‘like’, etc.15 Other explanations have been proposed, and there is no consensus among specialists about the etymology of -ként, but the hypothesis of an etymological meaning ‘in the shape of ’ is supported by the fact that this suffix has functive and similative as its two possible functions. 8.3 Verb ‘be like’ > similative marker > functive marker This grammaticalization path accounts for the development of the Zulu similativefunctive marker njenga- ‘like, as’ illustrated in Ex. (33) above. Njenga- consists of two formatives, nje ‘thus, in this manner’ and nga, reflex of the reconstructed Bantu verb *ngà ‘be like’.

15.  Note that a similar grammaticalization path is reconstructed for English like.



Functive phrases in typological and diachronic perspective 209

9. Grammaticalization paths resulting in the emergence of functive markers whose polysemy pattern includes locative and/or participant roles 9.1 Locative > functive A grammaticalization path locative > functive can be safely reconstructed for Finnish and Estonian. Apart from its use as a functive marker, the Essive case of Finnish is exclusively found in the expression of location, mainly temporal, but also marginally spatial. Moreover, this evolution is part of a more general shift that affected the whole system of Finnish spatial cases, involving the creation of two series of spatial cases (the in- and ad- series) and a tendency of the former spatial cases towards losing their spatial uses. Although de Groot (2010) states that the precise semantic motivation for this development is unclear, the hypothesis of an unmarked semantic type shift by which a concrete noun is converted into an abstract quality noun (‘in N’ being interpreted as ‘in the quality of N’) provides an explanation that is both intuitively satisfying and consistent with the general tendencies observed in semantic shifts. The functive use of French en (see Section 4.4 above) constitutes another clear case of a historical development from locative to functive. In the case of the West Mande language Soso, historical data about the postposition rá, used among others with functive and transformative noun phrases and in independent equative clauses, is of course not available, and the diachronic interpretation of the polysemy pattern of this postposition is not immediately obvious (since it includes both instrumental and spatial uses), but comparative data shows that it developed from a noun still attested in various Mande languages with meanings such as ‘place’, ‘side’, ‘opening’, ‘mouth’, and consequently grammaticalized as a spatial postposition before acquiring its other uses. Interestingly, even in cases of functive markers with a polysemy pattern that, at first sight, seems to have nothing to do with the expression of spatial relationships, it may happen that comparative data support the hypothesis of a locative origin. For example, the Mandinka postposition tí is productively used to mark functive and transformative noun phrases, and in the expression of the comparative of superiority; it is also marginally used to encode various participant roles (comitative, surrogative), but is never used to encode spatial relationships. However, comparative data from several other West Mande languages strongly support the hypothesis of a locative origin: – Bozo has a postposition ni, probably cognate with Mandinka tí, productively used (among others) in functive-transformative and spatial function; – Soninke has a postposition di exclusively found in spatial uses; – a noun tí glossed as ‘place’ is found in Bobo.

210 Denis Creissels

Consequently, the synchronic situation of Bobo, Soninke, Bozo, and Mandinka can be interpreted as representing four successive stages in the following evolution: a noun meaning ‘place’ (a) grammaticalizes as a spatial adposition, (b) acquires non-spatial functions (including functive and transformative), and eventually (probably due to the development of ‘younger’ spatial adpositions) (c) loses its spatial uses. Of course, this does not solve the problem of establishing how exactly the functive use may appear in this kind of evolution. Nothing ensures that the functive meaning developed directly from the locative meaning. In the particular case of Mandinka tí, it is interesting to observe that with one verb (sǎŋ), tí expresses surrogation: x y sǎŋ z tí ‘x exchange y for z’. Given the observations presented in Section 9.3 below, it is tempting to consider this surrogative use of tí as the vestige of a formerly productive meaning from which the functive use might have developed (rather than directly from the original spatial meaning). But at the same time, Mandinka has a verb yelemá interpreted as ‘pour (a liquid into a recipient)’ in combination with the locative postposition tó, but ‘transform (something into something else)’ in combination with tí, which might rather suggest a development of the type presented in Section 9.2 below, since Mande spatial postpositions are not specified for the distinction between static location and destination of movement. 9.2 Lative > transformative > functive A grammaticalization path lative > transformative > functive can be safely reconstructed for Mordvin. The functive marker -ks of Mordvin illustrated in Ex. (31) is also used as a transformative marker and is analyzed historically as a reflex of the Finnic-Mordvin Translative *-ksi. Comparative evidence supports the hypothesis that *-ksi descends from earlier lative suffixes (Grünthal 2003: 186), and this is consistent with the observation that, cross-linguistically, cases primarily encoding the destination of movement are very commonly used as transformative markers (Creissels 2008). 9.3 (Lative >) benefactive > surrogative > functive The formal similarity between two Basque cases (the so-called Benefactive and Prolative) specialized in benefactive and predicative (including functive) functions respectively, and the possible neutralization of their opposition with personal pronouns have been briefly presented in Section 5.2. This data strongly supports the hypothesis that these two cases have a common origin, and therefore that a case form or adposition expressing plain benefaction first extended its use to the

Functive phrases in typological and diachronic perspective 211



expression of surrogative benefaction, and subsequently acquired predicative uses. This hypothesis is consistent with the possible analysis of the ending -t of the -tzat formative common to the Benefactive and Prolative endings as cognate with a -t formative that can be added to the Lative ending -ra in Eastern Basque dialects (Joseba Lacarra, p.c.). 9.4 Separative > functive As already mentioned in Section 5.4, in some Romance languages, the reflexes of the separative preposition de ‘down from’ have acquired a functive use, but the historical processes that led to the development of this use are unclear, and the data I have been able to gather includes no clear case of a relatively direct development from separative to functive. 10. Conclusion This chapter summarizes the first results of a cross-linguistic investigation of the encoding of a semantic role whose importance and specificity have largely passed unnoticed so far. I only hope I have shown that the recognition of a distinction between participative, circumstantial, and predicative roles and a systematic study of functive and other non-participative roles may reveal interesting connections with the participative roles on which studies of semantic roles have focused so far, and consequently provide interesting insights into the conceptualization of events and its linguistic correlates. The data I have been able to gather shows that functive markers are commonly used to express other varieties of equative, and confirms that functive-similative syncretism is particularly widespread cross-linguistically. However, functive markers are also commonly involved in the expression of the following meanings: – – – – – – – –

transformative, comparative of superiority location, manner, instrumental, substitutive, lative, separative.

As regards the grammaticalization paths resulting in the emergence of functive markers, the situations in which the expression of the functive is one of the

212 Denis Creissels

possible uses of a case marker or adposition are particularly difficult to interpret, since the functive is typically expressed by ‘old’ case markers or adpositions whose original meaning is not always easy to reconstruct, and for which, even in cases when the original meaning is known or easily reconstructable, the reconstruction of the development of the functive use is problematic. However, in addition to the well-attested grammaticalization path deictic manner adverbs or words expressing similarity in complex constructions > similative markers > functive markers, the following grammaticalization paths have been identified: – – – – – – – –

‘in the quality of ’ > functive marker, dependent form of a copulative verb > functive marker, ‘in the way / state / shape of ’ > functive marker, ‘be like’ > similative marker > functive marker, locative marker > functive marker, lative marker > transformative marker > functive marker, benefactive marker > surrogative marker > functive marker, separative marker > functive marker.

Acknowledgements The following persons have contributed to this chapter by providing me with data, or helping me to orient myself in questions I am not familiar with: Ayten Babaliyeva (Tabassaran, Lezgi, Azeri), Mehmet-Ali Akıncı (Turkish), Sokhna Ba-Diop (Wolof), Alain-Christian Bassène (Jóola Banjal), Bertrand Boiron (Hungarian), Nthatisi Bulane (Southern Sotho), Jean-François Chaléat (Japanese), Claudine Chamorreau (Purépecha), Hilary Chappell (Chinese dialects), Injoo ChoiJonin (Korean), Michael Daniel (Archi), Danh Thành Do-Hurinville (Vietnamese), Gérard Dumestre (Bambara), Dmitri Ganenkov (Dargi), Csilla Kász (Arabic), Victoria Khurshudian (Eastern Armenian), Jérémie Kouadio (Baule), Joseba Lacarra (Basque), René Lacroix (Laz), Timur Maisak (Agul), Solmaz Merdanova (Agul), Zarina Molochieva (Chechen), Céline Mounole (Basque), Claire Moyse (Xaracuu, Wallisian, Futunan), Irina Nikolaeva (Nenets), Hayat Omar (Amharic, Oromo), Daniel Petit (Indo-European), Françoise Rose (Émérillon, Mojeño Trinatario, and other Amazonian languages), Wu Tong (Mandarin Chinese), Manana Topadze (Georgian), Nicolas Tournadre (Tibetan), Martine Vanhove (Maltese). This version has also benefited from Ekkehard König and Wolfgang Schulze’s comments on previous versions, and (last but not least) from the comments of two anonymous reviewers.

Abbreviations ABS = absolutive / ABSTR = abstract / ACC = accusative / ADV = adverbial / AOR = aorist / ASRT = assertive / AUX = auxiliary / CL = noun class / CMP = completive / COND = conditional / COP = copula / CVB = converb / DAT = dative / DEF = definite / DEM = demonstrative / EMPH = emphatic /

Functive phrases in typological and diachronic perspective 213



ERG = ergative / ESS = essive / excl = exclamative F = feminine / FOC = focalization / FUT = future / GEN = genitive / HAB = habitual / IMP = imperative / IMPF = imperfect / INCMP = incompletive / INDEF = indefinite / INESS = inessive / INSTR = instrumental / IPFV = imperfective / M = masculine / N = neuter / NEG = negation / NPST = non-past / OBL = oblique / PASS = passive / PFV = perfective / PL = plural / POS = positive / POSS = possessive / PRF = perfect / PROG = progressive / PROL = prolative / PRS = present / PST = past / REFL = reflexive / SG = singular / SUBJ = subjunctive / TERM = terminative / TOP = topic marker / TRANSL = translative

References Alekseev, Mikhail. 1988. Sravnitel’no-istoričeskaja morfologija avaro-andijskix jazykov [Comparative-historical morphology of Avar-Andic languages]. Moscow: Nauka. Aquilina, Joseph. 1999. English-Maltese dictionary, Vol. 1 (A-D). Valletta: Midsea books. Arkad’ev, Peter, Jurij Lander, Aleksandr Letučij & Nina Sumbatova. 2009. Vvedenie: Osnovnye svedenija ob adygejskom jazyke [Introduction: Basic information on the Adyghe language]. In Peter Arkad’ev, Aleksandr Letučij, Nina Sumbatova & Jakov Testelec (eds.), Aspekty polisintetizma: očerki po grammatike adygejskogo jazyka [Aspects of polysynthetism: Essays on the grammar of the Adyghe language], 17–120. Moscow: Rossijkij gosudarstvennyj gumanitarnyj universitet. Balogh, Judit. 2000. A névszóragozás [Noun inflection]. In Borbála Keszler. (ed.), Magyar grammatika [Hungarian grammar], 183–208. Budapest: Nemzeti Tankönyvkiadó. Bartels, Hauke. 2008. Sekundäre Predikation im Niedersorbischen. In Christoph Schroeder, Gerd Hentschel & Winfried Boeder (eds.), Secondary predicates in Eastern European languages and beyond (Studia Slavica Oldenburgensia 16), 19–39. Oldenburg: BIS-Verlag. Boeder, Winfried. 2005. Depictives in Kartvelian. Nikolaus P. Himmelmann & Eva SchultzeBerndt (eds.), Secondary Predication and Adverbial Modification: The Typology of Depictives, 201–236. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199272266.003.0006

Creissels, Denis. 2008. Predicative argument marking: The case of transformation verbs. Conference: Cognitive and functional perspectives on dynamic tendencies in Languages, Tartu. (Electronic version available at http://www.deniscreissels.fr/public/Creissels-pred. arg.pdf) Creissels, Denis. 2014. Functive-transformative marking in Akhvakh and other Caucasian languages. In Pamjati Alexandra, Evgen'eviča Kibrika, Vladimir Plungian, Michael Daniel, Elena Ljutikova, Sergej Tatevosov and Olga Fedorocva (eds), Jazyk, Konstanty, Peremennye, 430–449. Saint Petersburg: Aletja. Daget, J., M. Konipo & M. Sanankoua. 1953. La langue Bozo. Limoges. É. Kiss, Katalin, Ferenc Kiefer & Péter Siptár. 2003. Új Magyar Nyelvtan [New Hungarian grammar]. Budapest: Osiris Kiadó. Erelt, Mati & Helle Metslang. 2003. Case marking of the predicative in Estonian. Linguistica Uralica 39(3). 166–174. Ernout, Alfred & François Thomas. 1953. Syntaxe latine, 2nd edn. Paris: Klincksieck. Gimbatov, Magomed. 2006. Avar-ʕurus slovar [Avar-Russian dictionary]. Makhachkala.

214 Denis Creissels Groot, Casper de. 2008. Depictive secondary predication in Hungarian. In Christoph Schroeder, Gerd Hentschel & Winfried Boeder (eds.), Secondary predicates in Eastern European languages and beyond (Studia Slavica Oldenburgensia 16), 69–96. Oldenburg: BIS-Verlag. Groot, Casper de. 2010. Uralic Essive. Workshop ‘Uralic case’, 14th International Morphology Meeting. Budapest. Grünthal, Riho. 2003. Finnic adpositions and cases in change. Helsinki: Société finno-ougrienne. Hagège, Claude. 2010. Adpositions. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199575008.001.0001

Haspelmath, Martin. 1993. A grammar of Lezgian. Berlin & New York: Mouton de Gruyter.  doi: 10.1515/9783110884210

Haspelmath, Martin. 2010. Comparative concepts and descriptive categories in crosslinguistic studies. Language 86(3). 663–687.  doi: 10.1353/lan.2010.0021 Haspelmath, Martin & Oda Buchholz. 1998. Equative and similative constructions. In Johan van der Auwera (ed.), Adverbial constructions in the languages of Europe, 277–334. Berlin: Mouton de Gruyter. Hewson, John & Vit Bubenik. 2006. From case to adposition: The development of configurational syntax in Indo-European languages. Amsterdam & Philadelphia: Benjamins.  doi: 10.1075/cilt.280

Himmelmann, Nikolaus P. & Eva Schultze-Berndt. 2005. Issues in the syntax and semantics of participant-oriented adjuncts: An introduction. In Nikolaus P. Himmelmann & Eva Schultze-Berndt (eds.), Secondary Predication and Adverbial Modification: The Typology of Depictives, 1–67. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199272266.003.0001

Himmelmann, Nikolaus P. & Eva Schultze-Berndt (eds.). 2005. Secondary Predication and Adverbial Modification: The Typology of Depictives. Oxford: Oxford University Press.  doi: 10.1093/acprof:oso/9780199272266.001.0001

Holton, David, Peter Mackridge & Irene Philippaki-Warburton. 1997. Greek: A comprehensive grammar of the modern language. London: Routledge. Hualde, José Ignacio & Jon Ortiz de Urbina. 2003. A grammar of Basque. Berlin & New York: Mouton de Gruyter.  doi: 10.1515/9783110895285 Kuwae, Kunio. 1989. Manuel de japonais. Paris: L’asiathèque. Lafitte, Pierre. 1962. Grammaire basque (navarro-labourdin littéraire). Bayonne: Editions des “Amis du musée basque” et “Ikas”. Le Bris, Pierre & André Prost. 1981. Dictionnaire bobo-français. Paris: SELAF. Lelkes, István. 1967. Manuel de hongrois. Budapest: Tankönyvkiadó. Lutkat, Florence-Silvia & Cornelius Hasselblatt. 1993. Estnisch intensiv: das Lehrbuch der estnischen Sprache. Hamburg: Baltica. Magomedova, Patimat & Rašidat Xalidova. 2001. Karatinsko-russkij slovar’ [Karata-Russian dictionary]. Maxačkala: Dagestanskij Naučnyj Centr Rossiskoj Akademii Nauk. Mikkelsen, Line. 2011. Copular clauses. In Claudia Maienborn, Klaus von Heusinger & Paul Portner (eds.), Semantics: An International Handbook of Natural Language Meaning, 1805– 1829. Berlin: Mouton de Gruyter. Musaev, Magomed-Said. 1987. Slovoizmenitel’nye kategorii darginskogo jazyka (kategorii grammatičeskix klassov, abstraktnye padeži) [Inflectional categories in Dargwa (grammatical classes, abstract cases)]. Makhachkala. Nichols, Johanna. 1978. ‘Secondary predicates’. Berkeley Linguistics Society 4. 114–127.



Functive phrases in typological and diachronic perspective 215

Nichols, Johanna. 1983. ‘Predicate instrumentals and agreement in Lithuanian: A contrastive analysis’. Papers in Linguistics 16(3/4). 1–21.  doi: 10.1080/08351818309370593 Onikki-Rantajääskö, Tiina. 2006. Metonymy in locatives of state. In Marja-Liisa Helasvuo & Lyle Campbell (eds.), Grammar from the human perspective: Case, space and person in Finnish, 67–100. Amsterdam: Benjamins  doi: 10.1075/cilt.277.08oni Pinkster, Harm. 1990. Latin Syntax and Semantics. London: Routledge. Pinkster, Harm. 1991. Praedicativum. In Christian Touratier (ed.), Compléments prédicatifs et attributs du complément d’objet en Latin, 72–78. Aix-Marseille: Publications de l’Université de Provence. Rakhilina, Ekaterina & Elena Tribushinina. 2011. The Russian instrumental-of-comparison: Constructional approach. In Laura Janda & Marcin Grygiel (eds.), Slavic linguistics in a cognitive framework, 145–174. Peter Lang. Riese, Timothy. 1992. Der Translativ in den uralischen Sprachen 1. Linguistica Uralica 28. 251– 261. Riese, Timothy. 1993. Der Translativ in den uralischen Sprachen 2. Linguistica Uralica 29. 1–20. Rose, Françoise. 2011. Grammaire de l’émérillon teko, une langue tupi-guarani de Guyane française. Louvain: Peeters. Schönig, Claus. 2008. Notizen zu türksprachigen Gerundialformen (Konverbialformen) als verbale Kerne von Rollenprädikaten. In Christoph Schroeder, Gerd Hentschel & Winfried Boeder (eds.), Secondary predicates in Eastern European languages and beyond (Studia Slavica Oldenburgensia 16), 327–338. Oldenburg: BIS-Verlag. Schroeder, Christoph. 2008. Adverbial modification and secondary predicates in Turkish: A typological perspective. In Christoph Schroeder, Gerd Hentschel & Winfried Boeder (eds.), Secondary predicates in Eastern European languages and beyond (Studia Slavica Oldenburgensia 16), 339–358. Oldenburg: BIS-Verlag. Schroeder, Christoph, Gerd Hentschel & Winfried Boeder (eds.). 2008. Secondary predicates in Eastern European languages and beyond (Studia Slavica Oldenburgensia 16). Oldenburg: BIS-Verlag. Schultze-Berndt, Eva & Nikolaus P. Himmelmann. 2004. Depictive secondary predicates. Linguistic Typology 8(1). 59–131.  doi: 10.1515/lity.2004.004 Seki, Lucy 2000. Gramática do Kamaiurá, língua Tupi-Guarani do Alto Xingu. Campinas: Editora da Unicamp. Sohn, Ho-Min 1999. The Korean language. Cambridge: Cambridge University Press. Šarić, Ljiljana. 2008. Secondary predicates in Croatian. In Christoph Schroeder, Gerd Hentschel & Winfried Boeder (eds.), Secondary predicates in Eastern European languages and beyond (Studia Slavica Oldenburgensia 16), 295–325. Oldenburg: BIS-Verlag. Turunen, Rigina. 2011. Double encoding of nominal and adjectival predicates: A study of the nominative-translative switch in Erzya. SUSA/JSFOu 93. 385–415. Veksler, Bunim X. & Vladimir A. Jurik. 1975. Samoučitel’ latyšskogo jazyka. Riga: Zvaigzne. Vydrin, Arseniy. 2008. Are there depictives in Adighe? In Christoph Schroeder, Gerd Hentschel & Winfried Boeder (eds.), Secondary predicates in Eastern European languages and beyond (Studia Slavica Oldenburgensia 16), 423–445. Oldenburg: BIS-Verlag. Williams, Edwin. 1980. Predication. Linguistic Inquiry 12. 203–238.

Language index

A ASdyghe  198, 198n12, 213 Agul  174n1, 180, 205, 212 Ainu  33n7 Akhvakh, Northern  206 Amharic  174n1, 183, 212 Arabic  33n7, 174n1, 208, 212 Archi  174n1, 198, 202, 207, 212 Avar  191–193, 206–207, 213–214 B Balinese  33n7, 39, 39n11, 40 Bambara  174n1, 200, 212 Basque  19–20, 24, 174n1, 186, 188, 201–202, 202n13, 210–212, 214 Bemba  121, 125–131 Bezhta  33n7 Bobo  200, 210 Bora  33n7, 39, 39n11, 41 Bozo  200, 209–210, 213 C Chatino  33n7, 36–38, 41 Chechen  3, 54, 56–57, 59–62, 64–65, 77, 174n1, 188, 212 Chichewa  24, 76, 117 Chintang  32–33, 33n7, 76, 78 Choctaw  19, 25 Croatian  181, 215 D Dargi  174n1, 198, 206–207, 212 Djabugay  21, 25 Djambarrpuyqu  64 Dutch  129, 171, 207

E English  2, 4, 5n4, 6, 10–11, 11n6, 18, 28, 29n1, 32, 34n8, 43, 52, 54, 60, 98, 106–113, 115, 117–119, 121, 123–124, 128–129, 137n4, 139, 145, 155n10, 159, 163n12, 169, 175–176, 178–182, 186, 190, 194–196, 198, 200, 207, 208n15, 213 Erzya Mordvin  193 Estonian  19, 25, 185, 190, 192, 209, 214 Even  3, 11, 18, 33n7, 119, 128, 133 F Finnish  7–9, 11–13, 16–17, 79, 82, 85–88, 88n3, 89–91, 91n5, 92–103, 105, 107– 108, 118, 129–130, 157, 169, 179, 185, 192–193, 196, 197n11, 209, 215 French  157, 159–161, 164, 170, 187, 190, 195, 197, 207, 209 G Georgian  174n1, 197–198, 212 German  15, 22, 27–28, 32, 51–52, 117, 196 Greek  156, 194, 207, 214 H Hindi  181, 188 Hoocąk  33, 33n7, 39, 41 Hungarian  174n1, 185–187, 196, 208, 212–214

I Icelandic  32–33, 33n7, 34, 56n5, 76 Indonesian  33n7, 117 Italian  33n7, 137, 159–160, 162, 164–165, 170, 203 J Jaminjung  33n7, 39, 41 Japanese  116, 124n4, 174n1, 180, 182, 182n5, 212 Javanese  117 K Kamaiurá  186 Karata  180, 182, 214 Ket  33n7 Khanty, Southern  202 Korean  10, 83, 108, 116, 131, 174n1, 204, 212, 215 L Lardil  20, 25 Latin  23, 121, 133–134, 134n1, 135–136, 139–140, 140n7, 142, 142n9, 143–150, 154, 157–161, 164–168, 170–171, 197, 202–203, 207, 215 Latvian  196, 201 Laz  174n1, 188, 197, 212 Lezgi  174n1, 205, 212 Lezgian  7, 25, 83–86, 107, 169, 175, 214 Lithuanian  199, 215 Lower Sorbian  201 M Malay  33n7 Maltese  174n1, 208, 212–213 Mandarin  33n7, 111, 113, 174n1, 182, 212

216 Language index Mandinka  33n7, 176, 176n3, 177, 192, 196, 199–201, 209–210 Mapudungun  33n7, 116 Marathi  63, 77 Mari  202 Mordvin  186, 192–193, 210 N Nen  33n7 Nenets  174n1, 203–204, 212 Nias  64–65 O Old English  207 Old French  157, 159–160, 170 Old Italian  159–160 Oromo  174n1, 183, 212 P Persian  181 Proto-Tupi-Guarani  193 Purépecha  174n1, 186, 212

R Russian  21, 55n4, 90, 107, 184, 186, 190–192, 195, 199, 214–215

Tshangla  16 Tukang Besi  11, 24 Turkish  174n1, 183, 183n6, 206, 208, 212, 215

S Sinhala  117 Sliammon  33n7 Soninke  209–210 Soso  199–200, 209 Sotho-Tswana  183 Spanish  13, 116–117, 121– 122, 124, 128–129, 131, 159, 164, 180, 190–191, 195–196, 202–203, 207 Swedish  90, 95

W Wallisian  174n1, 181, 212 Warrungu  20 West Greenlandic  21

T Tashelhiyt  120, 131 Toba  121, 123–124, 124n4, 125, 128–130 Tsamai  57, 62 Tsez  64

X Xârâcùù  33n7, 39, 46, 174, 212 Y Yaqui  33n7 Yidiñ  17 Yoruba  33n7 Yupik  33n7 Z Zulu  195, 208

Subject index

A absolutive  20, 22, 24, 39, 53–54, 57, 61–62, 84, 213 accusative  27–28, 31, 133– 134, 136, 141n8, 147–148, 150–155, 158–159 actor  12, 27, 29, 189 see also agent addressee  134–137, 137n4, 142, 148–152, 154, 200 adelative  7, 9, 24, 83, 105 adessive  88n3, 105 adjunct  20, 116–117, 194 see also margins adposition  27–28, 92, 101, 141n8, 173, 175, 183–185, 189–191, 210–212 adverbial  4, 171, 195, 197–198, 205, 213–215 affectedness  4, 8, 15–16, 23, 82, 131, 142 agent  2–11, 14–24, 27–29, 37–41, 46–47, 52, 80, 82–86, 98–102, 105, 137, 151, 162–163 see also actor involuntary agent  82–83, 85, 107, 137, 162, 169 proto-agent  5, 7, 83 agreement  28, 32n6, 34, 129, 188, 215 alignment  27, 31–32, 39–42, 76–77 allative  7–10, 17, 53–54, 98–100, 105, 124, 129n5, 133–135, 157, 159, 161–163, 165–167 applicative  24, 116, 123–126, 130–131 argument  3–20, 27–34, 37, 51–62, 80–84, 92–93,

95–96, 98–104, 136–137, 139–149, 163–165, 180, 182 class  53–54, 57–59, 61–65, 67–69, 71–74 see also predicative argument attributive  186, 193 B Basic Linguistic Theory  4, 24, 80, 106 benefaction proper  109–110, 114, 116–118, 120–124, 127–129 benefactive  110–112, 114– 118, 120–121, 123–124, 124n4, 125, 127–128, 130– 131, 200–202, 210–212 beneficiary  8, 109–120, 123–125, 128–131, 137, 145–146, 162–163, 163n12 deputative beneficiary  15, 112–114 substitutive beneficiary  109, 114 C case frames  52, 54–57, 59–60, 62, 74–75 Case Grammar  2, 5 case marking  16–18, 20, 75–76, 83 grammatical case  18, 88, 91, 105, 133 non-default case  51–53, 57, 59–64, 67, 70, 73–74 semantic case  17–18, 138, 166 spatial case  24, 199, 209 causal  4, 24, 76, 81, 96, 201 causer  82, 118, 120, 137, 162–163, 167

cluster  27–28, 35–36, 38–41, 51–53, 55–57, 59–60, 65–67, 67n12, 68–75, 80, 137 see also role cluster cognizer  23, 51 comitative  80, 177, 199–200, 209 comparative  177, 196, 199, 201, 209, 211 Construction Grammar  5, 76, 79–81, 84, 106, 169 control  4, 8, 14–17, 40, 68, 99, 114, 116, 189 converb  130, 213 copula  90, 93n7, 179, 183, 189–191, 206, 212 D dative  59, 62, 120, 124–125, 133–134, 134n1, 135, 135n2, 136–137, 139, 142–162, 164–167, 186, 205 ethical dative  148, 158 depictive  183, 188–189, 214–215 destinative  201 ditransitive  55, 98, 157, 159, 163n12, 167 see also three-argument see also three-place E elative  88n3, 101–102, 105 equative  176, 176n2, 177–180, 182–184, 189, 191–193, 197, 200, 204, 209, 211 essive  88n3, 105, 173, 179, 185–187, 190, 193, 196, 197n11, 205, 208–209

218 Subject index experiencer  3–4, 16–18, 29–30, 51–52, 55, 59, 68, 71, 73, 85–86, 124, 143, 145 see also perceiver External Possessor  137, 145–146, 163, 167 F force  4–6, 14–15, 82, 84, 102, 117 functive  173–213 fuzzy cluster  23, 51–52, 65–67, 69–72 G genitive  24, 105, 159, 168, 201–202, 202n13, 213 goal  7–8, 12–15, 59, 72–73, 75, 86–87, 98–100, 134, 134n1, 135–137, 137n4, 140–142, 144–145, 148–159, 164–166 grammatical relation  19, 141–142, 146, 157, 159, 161–167 grammaticalization  98, 108, 131, 133–134, 166, 169, 173–174, 181–182, 204–205, 207, 207n14, 208, 208n15, 209–210, 212 I identification  91, 176–179, 184–186, 188, 190–192, 196, 199 illative  17, 24, 88n3, 92, 96, 99, 105 inessive  88n3, 93, 105, 205, 213 instigation  4, 14–15 instrument  2, 4–5, 11, 14–15, 18, 37–38, 40, 52, 72–73, 80, 84, 139–140, 173, 184, 205–204 instrumental  17–18, 21, 24, 138n5, 139, 164, 184, 186, 190, 195, 199–200, 205, 209, 211, 213, 215 intransitive  18, 31, 39, 41, 51–52, 55, 92–93, 117–119, 122, 124–125, 141, 189 see also one-argument see also one-place

L lative  62, 199, 202, 205, 210–212 Lexical-Functional Grammar  52, 76 Lexical-Functional Syntax  3 location  2–3, 10, 12, 18, 80, 126, 141, 148, 150, 186, 196, 202–205, 209–211 locative  17–18, 22, 24, 74, 76, 123–126, 129–130, 134n1, 202, 204–205, 209–210, 212 M macrorole  27, 29, 29n2, 30–32, 35, 41, 80 maleficiary  8, 125, 137, 145–146, 162, 166 manner  34, 82–83, 86, 89, 94, 96, 116, 140, 173, 194, 197, 202–205, 207–208, 211–212 margins  136, 136n3, 139–140, 143–148, 158, 162, 164–166 see also adjunct mesorole  29–30, 35, 41 metaphor  5, 84, 133–137, 141, 148–152, 154–159, 161, 163–166, 170 meteorological events  10, 83 metonymy  100, 103, 133, 135–136, 148–150, 154–155, 157–158, 166 microrole  27–33, 33–41, 43, 45, 47, 49 see also subrole monotransitive  14, 31, 123 see also transitive see also two-argument see also two-place N NeighborNet  23, 51–52, 66–67, 67n12, 68–72 nominative  20, 27, 32, 34, 39, 88, 90, 92, 99, 143, 159, 179, 182, 205 O object  2, 4, 6, 39, 55, 57, 92–96, 115–118, 133, 135, 138, 140–144, 146–147,

149–151, 153–154, 157, 159, 161–167 
oblique  4, 10–11, 18, 24, 55, 88–89, 95, 105, 119, 130, 141, 168, 176, 180, 189, 199 one-argument predicate  56–58 see also intransitive one-place predicate  52–53 see also intransitive onomasiological  55, 181 P partitive  13, 24–25, 85, 88, 94, 105 patient  2–7, 9–12, 14–18, 22, 27–29, 36, 38–41, 59, 80, 85–86, 110, 114–116, 118–123, 180 proto-patient  5 see also undergoer perceiver  23, 51–52, 73, 75 see also experiencer perlative  199 polysemy  31n5, 42, 92, 94, 104, 173–174, 185, 205–205, 209 possession  10, 58, 86, 90, 109–111, 111n1, 113–114, 116, 118, 129, 143, 168 possessor  3, 23–24, 51, 114–115, 117, 130, 137, 143, 145–146, 162–163, 167 predicate-specific roles  52– 53, 59, 61, 63, 69–70, 75 predicative argument  180, 213 see also argument prolative  186, 201, 210–211, 213 punctual coding  138, 140–143, 146–147, 150, 162, 164–167 purpose  2, 14–15, 23, 56, 81–82, 87n2, 122, 129, 131, 134n1 R recipient
  6–9, 15, 17, 36, 83–84, 86–87, 99–101, 105, 109, 111–115, 123, 125, 128, 129n5, 134–137,

Subject index 219

139, 142, 145, 150–153, 155–156, 162–164 Relational Grammar
  3, 31 relational coding  138, 141–142, 146–147, 157, 159, 162–163, 166–167 role accumulation  137, 144 Role and Reference Grammar  3, 52, 80, 112 role cluster  23, 27, 29, 31, 33, 35, 37, 39, 41, 43, 45, 47, 49, 51, 53, 55, 57, 59–61, 63, 66, 71, 73–75, 77, 137 see also cluster role complex  1, 23, 73, 133, 137, 137n4, 142, 146–147, 149, 159, 162, 164 S secondary predicate  173, 176n3, 183–184, 189–190, 192, 201, 206–207 semasiological  28 separative  202–205, 205, 211–212 serial verb construction  113, 117, 129, 182 similative  176, 176n2, 183–184, 186–187, 193–196, 204, 207–208, 211–212, 214 source  8, 12–13, 23, 51, 72–73, 75, 80, 84, 91, 101, 134n1, 150–151, 155, 157, 159, 161, 165 standard of comparison  177, 196 stimulus  3, 6, 17, 29, 36, 55, 85–86, 143

subject  4–5, 18, 32n6, 33–34, 39, 41, 57, 67–68, 138, 140, 143, 145–146, 162, 197–198 subrole  1, 14 see also microrole surrogation  23, 109–110, 113–114, 117–129, 210 syncretism  179, 183, 192–196, 199–202, 211 T thematic relation  14, 109–110 theme  3, 6–7, 10, 12, 14–15, 17, 52, 59, 71–73, 75, 84, 116, 118, 120 theta role  3, 79 three-argument predicate  56–59, 62, 68, 73 see also ditransitive three-place predicate  51–52, 57, 63, 74–75 see also ditransitive three-place verb  137, 142n9, 144, 147–148, 161–162, 164 see also ditransitive transfer  7–10, 17, 72–73, 75, 84, 105, 134–135, 150–151, 154–158, 165, 200 transformative  176–180, 189, 192–193, 197–200, 202, 204, 206–207, 209–213 transitive  16–18, 31, 39, 41, 51, 55, 83, 85, 90, 92–93, 118, 123, 164, 180, 189 see also monotransitive see also two-argument

see also two-place translative  88n3, 186, 193, 193n8, 210, 213, 215 two-argument predicate  51, 53, 56, 58–59 see also monotransitive see also transitive two-argument verb  31 see also monotransitive see also transitive two-place predicate  51–52, 56–57, 60, 63, 75 see also monotransitive see also transitive two-place verb  27, 133, 141–142, 144, 147–148, 162–165, 167 see also monotransitive see also transitive U undergoer  27, 29, 33, 80, 189 see also patient V verbless construction  79, 81, 87–98, 101, 103–105, 107–108 vocative  168, 205 volitionality  4–5, 15, 58, 84 Z zero-marking  9, 11, 188–189, 192

Especially in functional-typological linguistics, semantic roles have been studied thoroughly, because they constitute a good starting point for any study on argument marking due to their semantically deined nature. However, the very concept of semantic roles is far from being without problems, and there is still no consensus on how the roles are best deined. In this volume, the notion will be discussed from novel perspectives with the aim of providing new insights into our understanding of semantic roles. Two of the papers deal with semantic role clusters, one with semantic roles in verbless constructions, one with diachrony of semantic roles and two with individual semantic roles that have not been studied in too much detail in previous studies. The book may not ofer answers to all questions the readers may have, but at least it raises interesting further questions relevant to arriving at a better understanding of semantic roles. Originally published in Studies in Language 38:3 (2014).

ISBN

978 90 272 4276 1

John Benjamins Publishing Company