M.

Lemmens, Lexical Perspectives on Transitivity and Ergativity, [Current Issues in Linguistic Theory 166], Amsterdam & Philadelphia: John Benjamins, 1998. CAVEAT: the page layout of the present version is not fully identical to the original

CHAPTER 1
INTRODUCTION

1.1 Constructional variation with causative verbs This work ultimately argues that in order to explain the variable behaviour of causative verbs in modern English, it is fundamental to consider not only the lexical properties of those verbs but also the interaction between those properties and the meanings of the constructions in which the verbs may appear. My corpus-based analysis of verbs of killing, verbs which in linguistic treatises are generally taken as causative verbs par excellence, reveals the complex conceptual networks of (related) meanings instantiated by these verbs. With respect to constructional meaning, I continue, and by doing so refine, the view defended in Davidse (1991, 1992) that the English grammar of causative constructions is governed by the transitive and ergative paradigms. Overall, my perspective on the relevant grammatical constructions remains a cognitively inspired lexical one, aiming at clarifying how lexical and constructional meaning interact dynamically. The linguistic phenomena to be addressed in this book can best be identified via some seemingly puzzling questions related to expressions of killing: - Why can one rephrase John choked Mary (a causative) as Mary choked (a non-causative), while one cannot rephrase John killed Mary as *Mary killed ? - The same type of rephrasing is acceptable for drown in the sentence The former government drowned hundreds of dissidents (vs. Hundreds of dissidents drowned after being dumped into the ocean from a plane). However, it is not acceptable in The protesters horns drowned out the bell (*The bell drowned out). Why ? - Why does murder allow an objectless construction with the agent in subject position, as in, They have the clear intention to murder, whereas

M. LEMMENS in a comparable objectless construction with starve, e.g. They have the clear intention to starve, the subject cannot be interpreted as the agent causing someone elses starvation but as the entity that is affected by it ? - Why is it that many causatives can occur in a medio-passive construction, e.g. This book reads easily or These glasses break easily whereas such constructions are at least awkward for verbs of killing, e.g. *?Mary kills easily (with Mary being the victim). - Why is it that abort allows the causative/non-causative alternation in metaphorical uses (e.g. the pilot aborted the takeoff vs. the takeoff aborted), but does not allow it for literal uses (e.g. the woman aborted the child vs. *the child aborted) ?
LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY

While these questions are mere appetizers, they all relate to a verbs constructional variability coupled with its distinct meanings. To adequately answer these and related questions concerning the interaction of verb meaning and construc-tional variability, it is necessary to characterize verbs of killing against the background of the transitive and ergative paradigms. In contrast to, for instance, some Asian or Amerindian languages, English does not formally mark the ergativity/transitivity distinction (cf. Dixon 1979, 1994; Comrie 1981 or DeLancey 1981, 1984a, 1990 for illustrations of such markings). As a result, many scholars have failed to observe it as a principle operative within the grammar of English. However, can one claim at all that the English grammar of causation is structured by the ergative and transitive systems if the distinction between the two is not marked overtly, e.g. by case markers? Following Davidse (1991, 1992), I argue that the claim stands up to close examination, provided that one recognizes that overt markings are but one way in which the two systems manifest themselves in a language. I accept the Whorfian view that overt grammatical categories should be distinguished from covert ones which make themselves felt by the systematic relations that link all the paradigmatic correlates of a construction type. The semantics of a covert grammar can be arrived at by systematically examining the paradigmatically related constructions. This type of heuristics has been exploited insightfully within the framework of systemic-functional grammar as, for instance, in Davidses (1991) innovative work on transitivity and ergativity in the English grammar. Within mainstream linguistics Levins (1993) work on English verb classes is noteworthy. Levin effectively uses the alternation method to characterize a verbs meaning. The major shortcoming of her work is that she sees the choice

INTRODUCTION

of constructions in which a verb may occur as wholly determined by the verbs semantics and, as a result, fails to recognize that the constructions themselves are meaningful (see Chapter 2, section 2.2.3 for more explicit criticism). That constructions have meaning independent of verbs is argued, for example, in Goldberg (1995). While she discusses construction types different from the ones analysed in this work, her views are largely compatible with mine. Constructions are meaningful in and of themselves, but as Goldberg rightly insists, it is clearly not the case that the grammar works entirely top-down, with constructions simply imposing their meaning on unsuspecting verbs (1995:24). Rather, she says that the meaning of constructions and verbs interact in non-trivial ways (1995:24). The interaction between verbal and constructional semantics is indeed a complex matter. Within the scope of the present work, it can be restated in terms of the following questions: - How does the lexical content of a verb influence constructional variability? - Conversely, how do the various constructions in which a verb may occur affect its meaning? - What is the role of extension mechanisms like metaphor and metonymy with respect to the ergative-transitive interplay? - How may lexical and constructional properties change over time and does this lead to prototype shifts? My aim is to suggest some answers to these complex and far-reaching questions by analyzing a large corpus of verbs of killing. 1.2 Structural organization of this book 1.2.1 Part I: Theoretical Premises The first part of this book consists of three chapters which present a more general and theoretical exploration into some issues that are immediately relevant to my overall argument. As the first of these three chapters, the present chapter has stated the purpose of this book as essentially involving an account of constructional variability with lexical causatives. It further clarifies the general theoretical assumptions underlying this work (section 1.3), which are in essence an innovative synthesis of (predominantly) Cognitive Grammar and Systemic-Functional Grammar. The chapter concludes with a clarification of the strong empirical basis of our analysis (section 1.4). The next two chapters of Part I deal more elaborately with the grammatical and lexical perspectives which form the backbone of my work.

M. LEMMENS
LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY

Chapter 2 presents a careful description of causative constructions. First, I argue how lexical and analytical causatives, which have often been regarded as semantically equivalent, in fact involve quite different conceptualizations of an event (section 2.1). The second and most important part of Chapter 2, section 2.2, is concerned with a proper characterization of lexical causatives. It offers brief critical discussions of some views which have become accepted within certain schools of thought and which in one way or another are relevant to the theoretical synthesis underlying the present work. This synthesis, a cognitive lexical-paradigmatic approach, is characterized briefly in the last section of Chapter 2 (section 2.2.5); its descriptive and explanatory adequacy is fleshed out to the fullest in the descriptive analyses in Part II of this book. Chapter 3 takes up the other important pillar of my work, viz. the lexical semantic perspective on (causative) verbs. Clearly, the lexical perspective focuses on those aspects of verbal meaning that directly pertain to the verbs constructional potential. The chapter clarifies the cognitive view on semantic multiplicity, which I hold as the most appropriate to account for semantic variability both in lexis and in grammar. Following Langacker (see References), it is argued that conceptual categories are not rigidly and absolutely defined structures, but prototypically structured networks of interrelated meanings. It will be shown that this not only holds for lexical categories as for example instantiated by individual verbs of killing (section 3.1), but also for lexical fields, such as that designated by the group of verbs of killing (section 3.2). The next section in the chapter offers a preliminary clarification of the relationship between constructional variability and lexical meaning relatedness. The concluding section in Chapter 3 is in fact the conclusion to the whole of Part I (note that neither Chapter 1 nor Chapter 2 have a conclusion) and is at the same time an opener into the descriptive analyses in Part II. 1.2.2 Part II: Case Studies Part II of this book, comprising the actual descriptive work, contains four chapters which have been set up to cogently describe the complexities of the transitive and ergative models and their dynamic interaction with verbal meaning. Chapter 4 focuses on how the lexical structure of verbs may either constrain or encourage constructional extensions. Following a brief introduction, the first analysis (section 4.2) deals with lexically determined constraints on the middle or medio-passive construction in present-day English. Ironically perhaps, the insights on the middle construction originated in the need to account for the unmistakable absence of this construction for verbs of killing. The description augments Langackers cognitive analysis by a paradigmatic perspective, yet at

INTRODUCTION

the same time remedies some shortcomings of Davidses paradigmatic views on this construction. The second description offered in Chapter 4 (section 4.3) is a diachronic perspective on different types of paradigmatic shifts within the field of killing. It shows how over a considerable time span, the lexical evolution of some verbs, e.g. starve or throttle, has taken the verbs into new constructional possibilities. Chapter 5 deals with the Agent-centredness of the transitive paradigm, and explores this characteristic from different perspectives. First, it considers the prototype structure of the Agent-category and shows how agentivity is experientially motivated (section 5.1). Next, reviewing the individual MURDER verbs, it reveals how the Agent-centredness of this most salient (and prototypically transitive) subset of the field of killing is lexically determined (section 5.2). In a lexical-paradigmatic analysis of agentive nominals in -er (e.g. killer), section 5.3 explores the Agent-centredness of the transitive paradigm from a morphological perspective. Finally, our analysis of the transitive construction with omitted object further elucidates, from a constructional point of view, the fundamental hook-up with agentivity that characterizes the transitive paradigm. As a logical complement to Agent-centredness of the transitive paradigm, Chapter 6 shows how the ergative paradigm is primarily concerned with the participant affected by a process. By discussing a group of predominantly ergative verbs, the SUFFOCATE verbs, the chapter further reveals the experiential basis of ergativity for these verbs, yet it also shows how the opposition between transitive and ergative SUFFOCATE verbs is experientially motivated as well. The rest of the chapter reviews the enormously rich semantic and constructional coverage of the SUFFOCATE verbs. Section 6.2 elaborates the brief diachronic description of Chapter 4, section 4.3.2, and further substantiates the lexical and constructional overlapping that characterizes the history of these verbs. Next, section 6.3 (which forms the logical counterpart to section 5.2 in Chapter 5) describes the reorientation towards more differentiated lexical and construc-tional prototypes underlying the evolution to the present-day situation. The final section of the chapter discusses specific cases of transitivization of ergative choke and drown triggered by particles. Chapter 7 provides a maximal focus on the complexities of the interplay between the transitive and ergative paradigms, elaborating on the paradigmatically mixed character of the verb abort. By tracing the complex etymological evolution of this verb, we can account for the verbs constructional complexity in present-day English. It finds a logical explanation in view of the symbiosis of lexical and constructional meaning.

M. LEMMENS
LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY

By looking at grammatical structures from the lexical end, the different chapters in Part II clarify how transitivity and ergativity operate within the field of killing and as such may provide a useful basis for extrapolations to other types of causative verbs in the English lexicon. Starting from a general overview of my findings, the final conclusions to this work further reflect on the nature of verbal and constructional interaction and sketches some avenues for further research. To facilitate the comprehension of the text, an index and a glossary of terms are added at the end of this book. The glossary briefly defines the terminology used in this work and presents a list of corpus references used (see also section 1.4 below). 1.3 Theoretical framework As to its underlying assumptions, the present work adheres primarily to the principles of Cognitive Grammar as developed by Langacker (see References), especially in its treatment of lexical meaning. In addition, this study draws considerably on Systemic-Functional Grammar as developed by Halliday and others (especially Davidse 1991) whose views are in many respects strikingly compatible with those of Cognitive Grammar. At the same time, this work incorporates some descriptive insights from work in the framework of Generative Grammar and Relational Grammar and their derivatives (e.g. Keyser & Roeper 1984, Levin 1993; Levin & Rappaport Hovav 1995). As such, the theoretical underpinnings of this work offer what I believe to be an innovative synthesis of these frameworks. While the result may no longer be fully com-patible with rigid formulations of any of these, it offers insights which can be pertinent to all. The following sections briefly map out the major assumptions of these frameworks. The discussion, necessarily very brief, starts with Cognitive Grammar, the theoretical homeground of this book. Next, it discusses some notions of Systemic-Functional Grammar from which the insights concerning the paradigmatic opposition between transitivity and ergativity are borrowed. Finally, some aspects of Generative and Relational Grammar deserve some attention. It should be stressed that these descriptions serve to situate the present work in the linguistic tradition and do not aspire to be a full-fledged evaluation of these models. More critical evaluations occur at various places in this book, when specific views (e.g. on analytical versus lexical causatives or on agentivity) are at issue. 1.3.1 Cognitive Grammar Within the framework of Cognitive Grammar, there is no principled distinction between lexicon and grammar, both of which form a continuum of

INTRODUCTION

symbolic structures. More precisely, grammar is defined as a structured inventory of conventional linguistic units, i.e. structures which the speaker has mastered thoroughly to the extent that he can employ [them] in largely automatic fashion, without having to focus his attention specifically on [their] individual parts or their arrangement (Langacker 1987a:57). Three types of units are distinguished: semantic, phonological, and symbolic. The latter are bipolar and as such associate a phonological structure with a semantic structure. To say that lexical and grammatical structures form a continuum does not mean that they are identical; grammatical structures are more schematic than lexical ones. A sentence like The dog / bit / the postman combines three meaningful units into a larger structure which itself is also meaningful as it represents the sequence AGENT-PROCESS-AFFECTED.1 What this grammatical structure means is that some Agent acts upon another entity, the Affected. In this particular example, the identity of the Agent and the Affected are specified by the lexical items dog and postman, and the action by bite. The definite articles and the tense and voice of the verb are also regarded as meaningful. Langackers symbolic units are equivalent to the constructions in Construction Grammar (see e.g. Fillmore 1986, 1988; Fillmore et al. 1988; Kay 1990 or Goldberg 1992, 1995). Like Cognitive Grammar, Construction grammar does not assume a strict division of lexicon and syntax. As Goldberg puts it, both lexical and grammatical constructions are essentially the same type of declarative data structure: both pair form with meaning (1995:7). I will be using the term construction more restrictively, viz. in reference to the grammatical frames in which a verb may occur. For instance, the government starved its own children realizes an ergative effective construction (INSTIGATOR-PROCESS-MEDIUM), whereas the Ethiopians starved realizes the ergative non-effective construction (MEDIUM-PROCESS). These two constructions are formally and semantically different: the effective one overtly codes the entity that initiated the given event, while the non-effective remains neutral as to whether it was externally or internally instigated (see Chapter 2, sections 2.2.4 and 2.2.5, for more details). Returning to Cognitive Grammar, symbolic units are seen as providing the means for expressing conceptualizations in linguistic form. Cognitive grammar thus equates meaning with conceptualization. That is, semantic structure is defined as conceptualization tailored to the specifications of linguistic convention (Langacker 1987a:99). The meaning of a linguistic expression is a cognitive structure characterized relative to cognitive domains, where a
1

In fact, the nominals the dog and the postman are themselves already complex units, but I decide to ignore this here.

M. LEMMENS 8 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY domain can be any sort of conceptualization: a perceptual experience, a concept, a conceptual complex, an elaborate knowledge system, etc. (Langacker 1991a:3). As Lakoff (1987) has shown, most of these are idealized cognitive models (ICMs). Such models are similar to Fillmores frames, defined as unified frameworks of knowledge, or coherent schematizations of experience (Fillmore 1985:223). In this view, linguistic semantics is encyclopedic in nature, involving specifications in many cognitive domains.2 The traditional distinction (cf. Lyons 1977) between linguistic knowledge and encyclopedic knowledge is thus untenable. As Haiman (1980:331) states it, dictionaries are encyclopedias (cf. also Langacker 1987a:154ff, Croft 1993). The relevant background information for the characterization of meaning constitutes a network of shared, conventionalized, to some extent perhaps idealized knowledge, embedded in a pattern of cultural beliefs and practices (Taylor 1989:23).3 Against the background of such larger conceptual structures, called the base, linguistic structures impose their own specifications, which brings us to one of the pivotal claims of Cognitive Grammar, viz. that linguistic expressions and grammatical constructions embody conventional imagery (Langacker 1988a:7). Meaning thus relies on our ability to conceptualize the same object or situation in different ways. As Casad summarizes it: the speakers ability to conceptualize situations in a variety of ways is, in fact, the foundations of cognitive semantics (1995:23). The following chapters will provide ample illustration of different dimensions of imagery; by way of illustration, I elaborate here only one, viz. figure/ground alignment. This dimension involves a basic principle of cognitive and perceptual experience by virtue of which humans perceive (or cognize) entities as standing out against others. The meaning of linguistic units can also be seen in terms of figure/ground alignment, viz. as the imposition of a profile on some relevant base. The profile is a substructure that is elevated to a special level of prominence within the base (Langacker 1991a:5). Different predications can be characterized relative to one and the same domain, yet as they normally
2

See Lakoff (1987:75ff) for an illustration of the complexity of conceptual matrices (or in his terms, Idealized Cognitive Models or ICMs). 3 Gruber (1985) insists on the distinction between linguistic and encyclopedic meaning, and in fact complicates matters by adding yet another, viz. conceptual meaning. In his opinion, lexical and conceptual meaning are essential parts of the meaning of a word or the criteria for its use [] Encyclopedic meaning is knowledge or beliefs merely associated with a word or category, i.e. inessential for its use (1985:255; my emph.). Gruber introduces these distinctions to explain degrees of semantic acceptability (or anomaly). These gradiences find a more natural account within an encyclopedic view on meaning as advocated here (see the cited works by Croft, Haiman, Lakoff, Langacker and Taylor.)

INTRODUCTION

diverge in their figure/ground alignment, they are semantically non-equivalent. As an example, consider the relationship between the verb kill and its nominalization killing in the following examples: (1) (2) Yusef Hawkins was killed last week (WSJ)4 the killing of a black youth by whites last month in Brooklyn. (WSJ)

Both examples refer to the same event, and from a truth-conditional perspective they would be considered semantically equivalent. The difference would be merely one of form. Within the framework of cognitive grammar, however, the verbal and nominal forms do not have the same meaning since they impose different profiles. They have the same conceptual base but differ with respect to how this base is treated. (I ignore some additional differences (e.g. the semantic import of the by and of-periphrasis) since they are not essential to the present discussion.) To appreciate the differences, it may be useful to take a closer look at profiles imposed by verbs and nouns. Verbs are relational predications that profile a temporal sequence of interconnections between entities. Langacker identifies such a state-by-state profile as sequential scanning. This close-up view stands in polar opposition to the holistic view an episodic noun imposes on the same base. Such a noun is characterized by summary scanning. It profiles the component states collectively as a thing, i.e. as a region in some domain, where a region is characterized abstractly as a set of interconnected entities (Langacker 1987a:214).5 In the case of an episodic noun, the process functions as the base for the predication: the temporal and relational dimensions are still present but are not in profile. Figure 1, based on Langacker (1987a:247), contrasts the profiles that define the two grammatical categories (where profile is indicated via the thickness of the lines).

4 5

The reference serves to identify the source of the example. See section 1.4 below for details. As is obvious, the term thing is by no means limited to physical objects, although the latter are probably typical instantiations of the category. The term is used as an abstract technical term to refer to the product of any conceptual reification. See Langacker (1987a) on the difference between sequential and summary scanning.

10

M. LEMMENS
LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY
event (verb) tr reification (nominalization) (a) tr tr (b)

lm

lm

lm

Fig. 1: Process (a) and reified process (b) Figure 1(a) represents a processual expression which profiles the relationship between the participants (represented by the two circles) in its evolution through conceived time (indicated by the heavy-line segment of the time arrow). Figure 1(b) represents a nominalization as a conceptual reification of an event conception (Langacker 1992:497). It profiles the abstract region of intercon-nected states. To keep things manageable, the simplified representation shows only three states of what in fact is an indefinite succession of states. Relational predications, central to this book, are more complex than nouns since they have a complex profile that also manifests figure/ground (= trajector/landmark) alignment. Within the relationship represented by Figure 1(a), the trajector (tr) is identified as the primary participant, the landmark (lm), as the secondary participant. The passive construction is thus characterizable as a figure/ground reversal the switching of what is foregrounded in a given scene with what is backgrounded (Emanatian 1993; cf. also Langacker 1987a:120ff). To return to the basic assumptions of Cognitive Grammar, the grammar of a language is, as said, a structured inventory of conventionalized linguistic units (Langacker 1991b:511). All grammatical structures, be they morphological, lexical or syntactic, emerge as inherently meaningful. For example, while many grammars regard particles as meaningless, the description of choke {back/down/off/up/out} and drown out in Chapter 6, section 6.4, demonstrates their semantic contribution to the clause meaning and their impact on the verbs constructional possibilities. The constructions themselves are inherently mean-ingful too. Constructionally related pairs like John killed Mary versus John killed or Burglars murdered Jill versus Jill was murdered by burglars are semantically non-equivalent because they serve to conceptualize the event differently. In plain terms, the cognitive dictum is that any difference in form entails a difference in meaning (cf. e.g. Wierzbicka 1988 or Langacker

INTRODUCTION

11

1985, 1990b). Naturally, a given reality can be conceptualized in different ways. Chapter 6 argues that the lexical and constructional overlapping in the group of SUFFOCATE verbs is in fact largely due to subtle variations in the conceptualization of what is ontologically quite similar. An important characteristic of Cognitive Grammar is the view on (semantic) categories as not always well-delineated and structured around a prototype, the conceptual centre of the category. A category may have peripheral members, which are defective in one way or another; often these peripheral members lie in the blurred transition area between two categories. Langacker analyzes the members of a category as nodes in a networks, linked to one another by various sorts of categorizing relationships (1991a:266). Of these relationships, instantiation and schematization are particularly relevant, as it allows one to unite under one more abstract and general structure, called a schema, various structurally more varied lower-level constructions, the instantiations. A more elaborate discussion of Langackers model of categories is offered in Chapter 3, section 3.1. 1.3.2 Systemic Functional Grammar In many ways cognitive grammar is compatible with Hallidays systemicfunctional grammar. A language, Halliday says, is a system for making meanings (1985:xvii). As in Cognitive Grammar, the term semantic is not reserved for the lexicon, but pertains to the entire system of meanings of a language, expressed by grammar as well as by vocabulary (1985:xvii). As also observed by Wierzbicka, in language, everything conspires to convey meaning (1988:2). Parallel to the cognitive grammar views on the conceptualization of reality, Halliday emphasizes that the semantic structures of a language enable us to think about our experiencethat is to interpret it constructively (1985:xviii). Semantic structures being lexical and syntactical, Halliday thus commits himself to the view that syntax is motivated by meaning. However, an important difference with cognitive grammar (and other frameworks) is that Halliday sees the motivation as a natural (i.e. symbolic) relation between meaning and lexicogrammar (lexis and syntax) and that is why language is learnable: [children] can make the link between the categories of the grammar and the reality that is around them and inside their heads (1985:xviii). This plausibility is for instance reflected in the fact that English has verbs and nouns, to match the analysis of experience into processes and participants (1985:xviii). This congruence with experience can, however, be overturned by a process of grammatical metaphor, via which categories can be cross-coded (e.g. a process represented by a nominal phrase). Against the background of the previous discussion of Cognitive Grammar, the compatibility between

M. LEMMENS 12 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY Hallidays and Langackers view will be appreciated. Instead of congruence, Cognitive Grammar uses the notion canonical conception which, too, is seen as having a strong experiential basis. While the difference between Hallidays tri-stratal view of the linguistic sign (meaning - lexicogrammar - phonology) and the bipolar symbolic units of Cognitive Grammar calls for more careful analysis, it will not be elaborated in the present work, as it does not hamper the successful integration of the descriptive merits of the two models. Following Hjelmslev, Halliday characterizes language as system and process (1985:xxii). The system is to be seen as a highly complex network of interrelated choices: it is a paradigmatic, non-linear meaning potential (Davidse 1987:46). With the notion language as process, Halliday refers to actual instances of language use (which he calls text), which is linear and syntagmatic (1987:46). The paradigmatic nature of systemic grammar means that describing something consists in relating it to everything else (Halliday 1985:xxvii; his emph.). It is precisely by tracing the paradigmatic organization of language (e.g. via the alternation method) that relevant insights can be obtained regarding the grammar of actions and events. Up to the present, linguistic analyses in Cognitive Grammar have continued to focus primarily on the syntagmatic side and neglected the paradigmatic; this work expands Cognitive Grammar to include the latter into its analyses. Clearly, the paradigmatic options are constrained, not only from the grammar itself, but also from the context and the culture (Davidse 1987:46). This view is quite compatible with the cognitive view on linguistic construal. Next to language as system, equal weight is given within SystemicFunctional Grammar to language as process, where process is seen as text, i.e. any actual instance of language use in context. As Halliday points out, one must carefully consider how the system engenders text and at the same time, text should be related to the system behind it, since anyone understanding the text does so only because they know the system (1985:xxii). The present work will also have a strong empirical basis (cf. section 1.4 below), which allows us to come to nuanced judgments on how the paradigmatic contrast between transi-tivity and ergativity operates within English. A strong concern with data (language in context), as opposed to idealized language systems, is also a feature of Cognitive Grammar, characterized by Langacker as a usage-based model. In such a model, next to a concern for general rules and principles, we must also give substantial weight to their arrays of conventional instantiations, investigating the actual extension of the patterns in question and the factors that influence it (Langacker 1991a:265).

INTRODUCTION

13

All in all, despite some fundamental differences between systemic and cognitive linguisticsnotably, Hallidays multi-layered approach to language structure and his emphasis on the interpersonal dimension of grammarand differences of perspective or terminology, there is considerable affinity between the Systemic-Functional Grammar and Cognitive Grammar. Exploring the meta-linguistic divergencies and similarities between Systemic-Functional Grammar and Cognitive Grammar will surely be a profitable enterprise; however, as these are not directly relevant to the present work, they will be disregarded. The present study tries to break through the metalinguistic shells of different frame-works and distil from them their descriptive value. 1.3.3 Generative and Relational Grammar The theoretical assumptions of Generative and Relational Grammar (and their derivatives) are radically different from those of Cognitive Grammar and Systemic-Functional Grammar. A major criticism against Generative and Relational Grammar is that they assume syntax to be autonomous from semantics. Hence, they often claim relationships between structures (e.g. active and passive or nominal and verbal constructions), but see these as semantically equivalent. In both Cognitive Grammar and Systemic-Functional Grammar, in contrast, semantics forms the basis for grammar. Langacker stresses that grammatical structure reduces to patterns for the structuring and symbolization of conceptual content, and that all valid grammatical constructs have some kind of conceptual import (1991a:338-9). Similarly, Halliday points out that paradigmatic grammars take semantics as the foundation (1985:xxviii). In this view, any structure is meaningful and consequently, the notion of meaningless structures (dummies or chmeurs) is rejected. As has already been emphasized above, the basic view is that any difference in form entails a difference in meaning. Logically, then, the division between syntax, morphology and lexicon is not strict, as all consist of meaningful units, admittedly of different levels of schematization (Cognitive Grammar) or scales of delicacy (Systemic-Functional Grammar). A systemic inspired point of criticism is that the Generative and Relational models are exclusively syntagmatic in orientation; they regard language as a list of structures between which relations may exist (hence the transformations or strata). Mostly, however, they fail to see the deeper system at work in the grammar and thus often make incorrect distinctions (if at all). With respect to the issues central to this book, they mostly miss the awareness of the transitivity/ergativity distinction. More specifically, they generally do not observe the distinction between a transitive two-participant construction (e.g. John killed Mary) and an ergative one (e.g. John suffocated Mary) nor do they distinguish between transitive and ergative one-participant constructions, e.g.

M. LEMMENS 14 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY Mary died vs. Mary suffocated (see Chapter 2, sections 2.2.4 and 2.2.5 on these distinctions). Another problematic issue is the view on language in terms of the distinction between competence and performance, which especially in the Chomskyan model figures prominently. Instead of positing a model which presupposes a hypothetical ideal speaker and which imposes the abstract structures of this ideal language onto language itself, Cognitive Grammar and Systemic-Functional Grammar start by observing language use itself, which they see as the analytic way to capture generalizations concerning the underlying semantics (cf. above).6 From the earliest formulations of his theory, Halliday has incorporated the social dimension into his model, stressing that language is not a self-sufficient entity, but that it has an essentially instrumental character: it is used inand indeed evolved to servehuman interaction (Davidse 1987:40). Cognitive Grammar, too, defends such a position. Despite these (fundamental) theoretical conflicts, much valuable research has been carried out within the Generative and Relational Grammar models. For example, the Unaccusative Hypothesis, as first introduced by Perlmutter (1978) and later taken over (and silently extended) by Government and Binding theory (e.g. Burzio 1986, Marantz 1984) has been concerned especially with different types of intransitives and drawn attention to certain differences, yet it is still flawed in many respects. Other work that has contributed to the insights of the present book are Horn (1980), Keyser & Roeper (1984), Levin (1993), Levin & Rappaport (1995). Levins (1993) work is particularly interesting in two respects: (1) she explicitly couples a verbs constructional behaviour with its meaning and (2) she uses the heuristic of alternation to arrive at a proper characterization of the verbs constructional potential. As will be shown (Chapter 2, section 2.2.3), there a still some problems with her approach, one of them being her too rigid view on categories. Levin & Rappaport (1995) offer a more nuanced view which is considerably more compatible with those defended in this book. A brief evaluation is presented in the final conclusions of this book. In short, the present work wants to offer a synthesis of especially Cognitive Grammar and Systemic Functional Grammar and I believe that both models
6

The generative quest for general, preferably even universal, principles of grammar has tended to discourage the analysis of real life data. As Gross (1976:861) observes, within the generative tradition, motivated by a desire to treat linguistics at an abstract level, work based on sentences [] has almost entirely vanished. In essence, what the generativists were after was an abstract linguistic theory, not a model of language usage.

INTRODUCTION

15

can greatly benefit from this synthesis. Although the tide is turning, Cognitive Grammar has had, since its earliest formulations in the late 70s, some bias towards lexical semantics (think only of the numerous preposition and metaphor studies, e.g. Lakoff & Johnson 1980, Brugman 1983, Hawkins 1984). This work hopes not only to emphasize that constructions themselves are meaningful units (a concern also central to e.g. Goldberg 1995), but also to offer new insights concerning the transitivity/ergativity distinction as operative in English grammar. Systemic-Functional Grammar, in turn, can profit from the lexical orientation of this work. In my view, too little attention has been paid within Systemic-Functional Grammar to the influence of the verbs lexical structure on paradigmatic choices. It is in my emphasis on the lexically determined structural variability that my cognitive homeground manifests itself most clearly. 1.4 Empirical methodology 1.4.1 Advantages of corpus-based research Next to the theoretical framework, the methodology for data gathering and analysis greatly determines the conclusions of any descriptive study. While I do not want to overemphasize the practical decisions underlying this work, it is warranted that some of the choices be clarified. A major feature of the present work is its empirical foundation. I am committed to the idea that in order to gain insight into the principles of grammar, a close examination of a vast amount of contextualized, preferably non-elicited, language material is essential, for several reasons. First, a corpus-based analysis avoids inaccuracies and overgeneralizations which one often finds in intuition-based studies. Secondly, it avoids the overconcentration on typical examples which unambiguously illustrate whatever one is trying to prove. Extensive corpus material can shed light on less prototypical cases, leading to a better characterization of linguistic reality. Thirdly, corpus analysis may bring to the fore subtle semantic differences that would otherwise have been overlooked, as the analysis of collocational patterns for the SUFFOCATE verbs will prove.7 Clearly, a corpus-based analysis faces some shortcomings as well. First, it cannot with absolute certainty confirm the unacceptability of some or other construction, exceptions being explicit metalinguistic comments such as John died Mary is an ungrammatical English sentence.8 Nevertheless, provided
7 8

See e.g. also Sinclair (1991) or Stubbs (1995) for such collocation-based analyses. See Chapter 7 for an example of such meta-linguistic comment. Note that the problem of attestation also applies to diachronic research: the absence of a certain structure from historical documents does not necessarily mean that it has never existed.

M. LEMMENS 16 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY that ones database is sufficiently large, the non-occurrence of a pattern may be held indicative of its status within the grammatical system. In addition, absence of certain constructions may trigger new insights into the semantics of the construction itself, as has been the case for our analysis of the middle construction which does not occur in the text corpora. A second drawback of corpus analysis is that there are practical limitations to the amount of material which can be analyzed within a certain time-frame. Although larger computerized corpora as well as more powerful tools for analysis are becoming more and more available, the work load is still considerable. As most corpus-based studies, the present work has also been delimited by what was practically feasible. Undoubtedly, such practical limitations prevent a study from reaching the ideal of maximal representativity. While I find the notion of representativity itself quite challenging (cf. also Stubbs 1995), I have taken care to select data of different types in order to reach an acceptable degree of representativity. By doing so, certain reliable conclusions can be drawn from the data, indicating major tendencies and evolutions. 1.4.2 Corpora consulted More specifically, the corpora consulted comprise text from different registers, different regions (chiefly British versus American English), and different periods. Table 1 gives a short overview of the major (electronic) corpora used.
Reference
WSJ

Short Description Wall Street Journal: computerized version of 3 months of the paper Leuven Drama Corpus: collection of 62 British plays from the late 60s and early 70s Contemporary American Short Stories, taken from journals (Athene, Intertext, Quanta) distributed via the Internet Collection of some fifteen 19th century novels, distributed via the Internet
CD-ROM)

Size 5,353,500 words 1,029,660 words 1,066,875 words

LDC

CASS

NOV

746,525 words 11,713 citations

OED

Oxford English Dictionary (both paper edition and

Table 1: Overview of corpora consulted

INTRODUCTION

17

Each of the corpora has its own specific character. The WSJ corpus, for example, stands somewhat apart because of its highbrow register and its journalese.9 Nevertheless, it has been a rich resource for the MURDER verbs, given the many reports on various types of killings, murders, assassinations, and the like. The fiction corpora, in turn, have proven most rewarding for the SUFFOCATE verbs. Moreover, in comparison to the WSJ they show greater constructional variation; the reason is probably that in the fiction corpora, especially the CASS corpus, the creativity has not been hampered so much by newspaper editors (cf. Bell 1991) or by the formal enterprise of regular literary production under the influence of style-editors and market-conscious publishing houses. The Novel Corpus and the Oxford English Dictionary have been used to give some of our analyses diachronic depth. As for the OED it should be noted that I have consulted both the paper edition and the edition on CD-ROM. The latter made it possible to obtain any citation from anywhere in the dictionary. A specific reference to the CD-ROM version only occurs when the citation is not listed under the items lemma in the paper edition. Whenever relevant, OED citations carry the date of attestation. When the citations deviates strongly from contemporary English, a gloss will be added. It is stressed that this gloss tries to reflect as much as possible the original phrasing, and no attempt is made to render a polished translation. The strong data-oriented character of the present work is reflected in the consistent use of (mostly corpus-based) examples to support my claims. It has also been my policy to vary the examples as much as possible. At all times I refer to the source of the example, using the abbreviations specified above. Examples drawn from other linguistic treatises will have a reference to these works. At times, I bring into the discussion some occasional examples, i.e. ones that I have come across in my personal reading, that I heard on radio or television, or that I have taken from dictionaries. The reader is referred to the glossary at the end of the book for an overview of these occasional references. Examples without reference are my own or are self-made variations on attested usages. It should be noted that to enhance the readability of the text the examples have been slightly edited. Mostly, this involved leaving out irrelevant information. As a rule, the spelling has not been altered (e.g. the difference between American versus British spelling has been preserved). The database on which the present study is based has been arrived at by excerpting (via simple pattern-matching procedures) from the corpora all occurrences of kill-expressions. Chapter 3, section 3.2 gives a more elaborate
9

I kindly acknowledge the Association of Computational Linguistics for granting me permission to use the ACL WSJ corpus.

M. LEMMENS 18 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY and semantically based overview of expressions that are regarded as belonging to the field of killing. The excerpted attestations were subsequently marked with special codes to allow further analysis. It should be noted that the present work does not discuss all the extracted items in full detail. For instance, verbs of killing derived from instruments, such as to knife, to (machine-)gun or to poison are too few in number to yield significant conclusions. Expressions from semantically related fields, like die, lose ones life, perish, hunger or thirst are brought in for contrastive purposes. An overview of the items more elaborately discussed in Part II is presented at appropriate places in these descriptions. It should be noted that the extraction (and subsequent analysis) has not been restricted to verbs, but includes derived forms (e.g. killing, the kill, stranglehold, murderer, etc.) as well as related forms which are not directly derived from the verb (e.g. abortion). In total, the number of attestations excerpted from the text corpora amounts to approximately 4000. Augmented with those from the OED, the total of attestations examined adds up to over 15,000. The empirical basis of the present work has allowed us to come to more refined pronouncements on verbal and constructional meaning. As said, such a bottom-up approach furnishes insights into the complexities of lexicon and grammar from which general patterns and tendencies may be discerned. Clearly, while these generalizations are important, they are not the sole purpose of our investigation nor can they be regarded as telling the whole story in and by themselves. That is, they do not function as general rules from which all possible and well-formed linguistic structures can be predicted. Following Langacker (1987a, 1990b, 1991a), we reject such a one-sided quest for absolute predic-tability, which has figured prominently in Generative circles. Instead, we assume that abstract rule schemata coexist, and interact with, conventionalized instantiations. As Croft (1994:29) puts it, [general] explanations are welcomed where found, but language is a mixture of functional motivation and arbitrary convention. Our investigation does support some predictability, e.g. the strong Agent-centredness of the MURDER verbs predicts that an ergativization will be barred. Nevertheless, even for seemingly clear-cut cases, absolute predictability is to be rejected. For instance, the Agent-centred SUFFOCATE verbs strangle and throttle have, at one point, been subject to ergativization. Our analysis shows that unexpected phenomena like these can be motivated on cognitive grounds. They can be accounted for within our cognitive lexical-paradigmatic framework. In the following two chapters, the two pillars of this framework are described in more detail. Chapter 2 focuses on the semantics of causative constructions arguing for a

INTRODUCTION

19

paradigmatic approach. Chapter 3 will present our cognitive view on lexical categories. The two chapters will provide the background assump-tions which will be further substantiated in the descriptive analyses in Part II.

M. LEMMENS, LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY,

CHAPTER 2
THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

When talking about processes in general and causative processes in particular, linguists usually concentrate on the physical processes and very often on acts of destruction. A probable reason for this predilection is that these verbs express a straightforward kind of causation, in which a human agent directly affects a change of state in a secondary participant, the patient. As indicated by Lakoff, this type of interaction forms the prototype of causation, which is built into the grammar of languages (1987:55). In other words, verbs such as hit, punch, kill, or break are often foregrounded in linguistic treatises on causation, precisely because of their salient position in the conception of causation. In this chapter I will clarify my position vis--vis the linguistic tradition concerning the coding of processes and causation. In the literature, the discussion of lexical causatives has often included references to analytical causatives. While these are not immediately relevant to the descriptive analyses in this book, it is warranted that I briefly consider the differences between analytical and lexical causatives, as an introduction to the more elaborate analysis of the latter. Hence, in the first section of this chapter, I will take issue with the often-made claim that lexical and analytical causatives are semantically equivalent (cf. McCawley 1968, Dowty 1972, Hoekstra 1992) and demonstrate that it is linguistically inadequate and too much based on a universalist interpretation of extra-linguistic events. Even studies which tried to counter the alleged synonymy of lexical and analytical causatives (e.g. Fodor 1970, Katz 1970) are often preoccupied with the ontological situation. In their objections to the generative semanticists, most of these researchers draw on the criterion of the spatio-temporal unity of cause and effect that is supposedly implied by lexical but not by analytical causatives. This account runs into problems when it appears that certain lexical causatives can refer to a situation where cause and effect are not contiguous in time and/or space. In other words, both the defenders and the refuters of the synonymy between lexical and analytical causatives falsely assume, as DeLancey correctly observes, that anything and everything that can be said about a situation is part of the semantic represen-tation of any sentence which describes that situation (1991:345).

22

M. LEMMENS
LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY

I will argue that lexical and analytical causatives are semantically nonequivalent because they encode different conceptualizations of a given situation. Similarly, construal differences within the class of lexical causatives are to be regarded as reflecting differences in conceptualization. Evidence unequivocally shows that there are two types of lexical causatives. But what exactly is the conceptual difference between these two types of lexical causatives? Following Davidse (1991, 1992), I will argue that the difference reflects the distinct transitive and ergative systems. I will trace the increasing sensitivity in the linguistic tradition to the importance of transitivity and ergativity in the gram-mar of English. My own position synthesizes pertinent insights from predominantly Cognitive Grammar and Systemic-Functional Grammar (next to other sources). This cognitive lexical-paradigmatic approach is only briefly described in this chapter (section 2.2.5) as it will be amply illustrated in the descriptive analyses (Part II) which show how this approach adds a new and revealing dimension by exploring how the more general grammatical principles are influenced by the semantic structure of the verbs that instantiate them. 2.1 Analytical versus lexical causatives In various descriptions of causation, the verb kill has received prominent attention especially in view of its often defended, yet equally often refuted, synonymy relation to cause to die. Since it is also the item central to my data analysis, the following discussion will take kill as the starting point. According to Wierzbicka (1980) the characterization of kill in terms of cause to die has a die-hard tradition (she traces it back to the 17th century grammarian Wilkins) but, undoubtedly, the most well-known defenders of this characterization were the generative semanticists. Their analysis of kill postulated a decomposition into semantic primitives: they argued that kill derived (via lexicalization) from an underlying structure cause to die or cause to become not alive. The same deep semantic structure was assumed to underlie cause to die, which accounted for the synonymy between the two surface expressions (see in this respect McCawley 1968 and Lakoff 1970).10 The generative analysis of kill in terms of [Cause[Become[Not Alive]]] has received severe criticism. The most well-known reactions are probably Fodor (1970), Katz (1970), Shibatani (1972, 1973, 1976) and Wierzbicka (1980). These reactions counter McCawleys analysis, yet are still inadequate in their account of the complex meaning structure of kill. Furthermore, while all of these studies try hard to pin down the

10

In this generative analysis, the cause in the deep structure is an abstract verb which, although part of the lexical deep structure of the English verb cause, is not to be equated with it (cf. Dowty 1972).

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

23

difference between lexical causatives (like kill) and analytical causatives (like cause to die), they overlook or do not account for the two-fold distinction that characterizes the class of lexical causatives itself, represented, for example, by kill and break. In this section, I will reproduce some of Fodors, Katzs and Wierzbickas criticisms and discuss the remaining inaccuracies in their positions. Fodor (1970) gives the following reasons for arguing against deriving both the lexical causative kill and the analytical cause to die from an underlying composite structure cause to die. First, if both cause Mary to die and Mary die are constituents in the deep structure of John killed Mary, then it should be possible, Fodor says, to have an anaphoric do so-construction in which do so replaces the VP in the constituent sentence Mary die (1970:431). But, as his examples clearly show, while the do so-construction is possible with the analyti-cal causative, it is impossible with kill11: (1)a. John caused Mary to die and it surprised me that he did so. (Fodor 1970) b. John caused Mary to die and it surprised me that she did so. (2)a. John killed Mary and it surprised me that he did so. (Fodor 1970) b. *John killed Mary and it surprised me that she did so. Secondly, Fodor observes that in the case of the analytical cause toconstruction, one can cause an event by doing something at a time which is distinct from the time of the event (1970:433), whereas such a separation in time and/or space is not possible for lexical causatives: (3)a. John caused Bill to die on Sunday by stabbing him on Saturday. (Fodor 1970.) b. *John killed Bill on Sunday by stabbing him on Saturday. On the basis of this, Fodor characterizes analytical causatives as depicting compound and lexical causatives as depicting atomic causal events. Although there is a lot to be said for Fodors distinction between atomic and com-pound events, it appears that Fodor is referring to referential time, i.e. the temporal sequencing as it unfolds in the world. If this is so, his account fails to explain why it is nevertheless in principle possible, and unproblematic, for kill to be used in cases where referentially the kill-component is separated from the die-component. Let us consider some examples from our corpora. The first example is an excerpt from Ustinovs play The Unknown Soldier and
11

For reasons of clarity I have emphasized the relevant participants that figure in the individual clauses.

M. LEMMENS 24 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY his Wife (LDC), a conversation between General Maximus (G) and his Archbishop (A): G: A: G: A: G: A: G: A: G: Aah! (He runs the archbishop through with his sword). (Falling to his knees with an echo of the general' s shout) Aaah!Thank you, Maximus (Aghast) What have you made me do? I have taught you remorseI have given you a conscience as a parting gift Why did you make me kill you? Because you are weak, my son. Theres no weakness more pernicious than the weakness of the strong. (Examining the archbishop as though he were some curious, half-dangerous animal) Why are you smiling? (Begins to laugh very softly, and shakes his head feebly) What a question. (Slow) Do you derive pleasure from suffering? (No reply. The general loses his temper. He stabs the archbishop again and again.) There! There! Enjoy yourself! (The archbishop rolls over and dies. The general stands there for a moment, panting. He returns his sword to its scabbard.)

Although the use of kill is perhaps somewhat peculiar (but well-suited for dramatic purposes), it is still quite plausible, despite the temporal separation between the first lethal thrust of the Generals sword and the Archbishops dying (admittedly, the time span is considerably less than that in Fodors example, but it is nevertheless irrefutably present). This example shows that something more than mere time matching is involved. More specifically, it is the conceptual integration of the two components rather than their referential simultaneity which determines the acceptability. The Generals use of kill is appropriate as it incorporates the subsequent dying of the Archbishop as part of his kill-action, i.e. his action (the thrusting of his sword) and the Archbishops dying are conceptually tied together as one event, regardless of whether the actual realization in the world out there might involve some temporal sequencing. In any case, the focus is on the Generals deed, as is also indicated by his questions What have you made me do?Why did you make me kill you? When you come to think of it, the use of kill in Ustinovs play may not be so unusual after all, since temporal sequencing is almost inevitably present in any kill-event (coded by kill or another verb of killing). The most unmarked construal of caused events is one in which the cause precedes the event, and

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

25

often the two are known not to fully coincide in time.12 For instance, in the sentence (4) They went straight from trains to showers under Zyklon B, which killed in minutes (WSJ)

the adverbial in minutes emphasizes the efficiency of the poison gas, yet also acknowledges that there is some time lapse, a few minutes at most, between the emission of the gas and the lethal result intended. The time lapse does not, however, affect the acceptability of kill. Finally, consider the following example which I heard on the radio. On March 23, 1994, the Mexican presidential candidate, Luis Colosio, was assassinated at a meeting in Tijuana. The following morning, the newsreport on the Flemish radio station (BRTN) opened by saying that the presidential candidate was shot dead (doodgeschoten) and then continued to specify that two gunmen had shot him in the head and chest. The mortally wounded victim was brought to the hospital, where he died of his injuries (for which the broadcaster used the expression gestorven, the past participle of sterven die). If we take Fodors observations as pertaining to referential time, the use of doodgeschoten would have been problematic since the man was not shot dead but mortally wounded and died later in the hospital. Yet, the use of doodgeschoten is quite natural and even well-suited to open the newsreport because it directly specifies the essence of the matter. At the risk of arguing the obvious, I maintain that lexical causatives such as kill almost invariably code events in which the agents deed and the patients affectedness fail to coincide referentially. However, unlike analytical causatives (like the cause to-construction), lexical causatives neutralize this difference and code the components as concurrent and integrated. If the examples above felicitously employ kill to code events in which the cause and effect do not coincide in referential time, why then is such integration impossible for Fodors example *John killed Bill on Sunday by stabbing him on Saturday? The sentence is unacceptable because the temporal framing is related to points of reference that are too explicitly differentiated, viz. Saturday and Sunday. Fodors observation may hold for the particular example, but he misses the point by interpreting the adverb matching referentially, as do many descriptions of causation.

12

A construal in which the result precedes the cause, which Talmy (1985b) has termed resulting-event causation is illustrated by sentences like The vase broke from a balls rolling into it or He died from a car hitting him (Talmys examples).

26

M. LEMMENS
LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY

Another line of criticism against McCawleys analysis is offered by Katz, who criticizes it as a false account of the meaning of the formative kill , (1970:253) since it does not make a distinction between direct and indirect causation. Katz is probably the first to counter the cause to die analysis on the basis of this distinction, which since then has figured prominently in the discussion of lexical and analytical causatives (e.g. King 1988; Lakoff 1987; Lakoff & Johnson 1980; Shibatani 1972, 1973; Smith 1978; Vendler 1984; etc.). Drawing on a Wild West anecdote, Katz argues that someone (the gunsmith) may be indirectly responsible for someone elses death (the sheriffs, shot by some outlaw), but that this does not entail that this gunsmith killed the sheriff. This line of criticism has affinities with Fodors observation about the possible spatio-temporal distance between the causation and the caused event in analytical causatives, since the directness of causation is in a way also a kind of distance, either referential or conceptual. In other words, the more remote the agents deed is from the effect (remote in whatever sense), the more contextual support is needed to justify the mental gymnastics of integration into a lexical causative. As Lakoff (1987:55) phrased it, the more direct the causation, the closer the morphemes expressing the cause and the result (cf. also Lakoff & Johnson 1980:Ch.20; Haiman 1983). However, this iconicity principle is not entirely accurate as it relies on the referential situation, whereas what is at issue is a difference of conceptualization of possibly one and the same reality. Sentences like (5) (6) (7) John killed Bill. John caused Bill to die. Bill passed away.

may all refer to the same event, yet differ in the way this event is conceptualized. A case in point is the following example: (8) I protested against Hitlers anti-Jew laws. He killed off six million of us! (LDC).

in which Hitler is coded as the agent directly responsible, despite the fact that in reality he was quite remote to the actual killing. Against the mainstream inter-pretation, I will argue, following Davidse (1991), that the iconicity principle has validity only if interpreted non-referentially, i.e. when the conceptualization is iconic for the semantics of the participant-process relation in the grammatical construal (cf. below). Another problem of the iconicity principle is that while it holds for kill, it does not hold for all lexical causatives. In line with Fodors criticism, Wierz-

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

27

bicka (1980:162) argues that a semantic component like somewhere is part and parcel of the semantic structure of a sentence like Peter killed the cat in the attic, because both killer and victim are always thought of as being in one place. Unity of place is not implied by Peter caused the cat to die in the attic. Thus, she finds in the meaning of kill two simultaneous events, one involving the doer and one involving the victim (1980:165), whereas cause to die allows for a separation of doer and victim. Attempting to shed light on the issue, she brings in other verbs, such as harden (as in John hardened the metal), solidify (as in John solidified the metal), and thicken (as in John thickened the sauce). In each case, she claims that two simultaneous events can be found: John is doing something and, at the same time, the metal is becoming harder or the sauce is becoming thicker. Her solution parallels Fodors analysis of the verb melt: one can cause an event by doing something at a time which is distinct from the time of the event. But if you melt something, then you melt it when it melts (1970:433) and this, he maintains, also applies, mutatis mutandis, to kill. Fodor and Wierzbicka are both correct in pointing out that the conceived spatio-temporal co-extension of the action and the (coming about of the) effect are central to the meaning of kill, but both incorrectly level it with verbs such as melt or solidify, which differ from kill in allowing a spatio-temporal separation, as already observed by Lakoff and Ross (1972). In their discussion of anaphoric island phenomena, they observe that for some causative verbs the spatio-temporal cohesion is not compulsory. Consider the by now famous examples (cf. also Shibatani 1973:289): (9)a. The scientist finally hardened the metal, but it took him a year to bring it about. b. *John finally killed Mary, but it took him six months to bring it about. Lakoff and Ross try to salvage the analysis of both verbs as causatives by postulating a principle that constrains anaphoric reference to cases where the embedded intransitive verb is morphologically related to the transitive verb. This is true for hardenintr/hardentr, but not for kill/die, and hence the unacceptability of (9b). The ungrammaticality, they conclude, cannot be taken as a counter-argument to the causative analysis, contrary to Fodors claim (Lakoff & Ross 1972:125). I agree with Davidse (1991) that the morphological constraint postulated by Lakoff and Ross does not really explain why a-type verbs allow the non-coextension whereas b-type verbs do not. The above discussion has clearly shown that lexical and analytical causatives cannot be put on a par, but encode a different conceptualization. As our main concern is with lexical causatives, the analytical causative

M. LEMMENS 28 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY constructions will fairly much be ignored in the present work (see Davidse 1991:65-77 for some further discussion). In the course of the discussion it has been noted that not all lexical causatives behave in the same way, as illustrated by the differences between kill and melt. Is this difference significant? Most of the cited studies tend to ignore the difference, yet it has not gone by unnoticed in (more) recent studies. Langacker (1991b:408) observes that
there are of course many verbssuch as kill, throw, and chopin which the notion of causation is inherent yet sublexical, in the sense of not being coded individually. Though still implicit, that notion is somewhat more visible to the analyst with verbs like break, open, and cook, for it constitutes the only semantic difference between their intransitive and transitive variants.

Langacker makes a distinction between verbs in which the cause is sublexical and those for which it stands out more. This distinction strikes at the heart of the contrast between the transitive and ergative paradigms operative in the English grammar. Since the two paradigms are pivotal to the analysis presented in this book, it is essential that they are more closely examined. 2.2 Transitive versus ergative lexical causatives The early generativists studies, as well as their critiques discussed above, neglect the distinction between transitivity and ergativity. Ironically, around the same time as McCawleys first analysis of kill in terms of cause to die, M.A.K. Halliday published a more refined view on transitivity in his three part exposition Notes on Transitivity and Theme in English 1-3 (1967-8). In these articles, he proposes a considerably more refined analysis of related one and two-participant construals in English, distinguishing between what at that stage he referred to as the effective and the descriptive systems. Davidse (1991, 1992) builds on Hallidays insights and interprets the effective and descriptive systems as paradigmaticallynot morphologicallyencoding the transitive and ergative principles that govern the grammar of actions and events. Since I will elaborate her insights towards the end of this discussion, I will refrain from going into Hallidays views here. Instead, I will start with a short description of transitivity and ergativity as used in traditional typology which will provide the background for the semantic reinterpretation offered in both Langackers cognitive and Davidses paradigmatic approach. 2.2.1 Traditional form-based transitive/ergative typology The traditional views on ergativity come from typological studies and are based on morphology. Ergativity is restricted to languages with ergative case marking or syntactic patterning, such as ergative co-ordination patterns in complex sentences in (some) Australian Aboriginal languages (see e.g. Dixon

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

29

1979, 1994, Plank 1979, and Comrie 1973, 1978, 1981 for descriptions within the traditional typology). A language is ergative if the subject of an intransitive clause is treated in the same way as the object of a transitive clause, and differently from the transitive subject (Dixon 1994:1). In contrast, in a nomina-tive/accusative system, the subject of transitives and intransitives are coded alike (nominative), whereas the transitive object is marked differently (accusative). The ergative system is illustrated by Dixons examples from Dyirbal, examples (10) and (11); the transitive, by his examples from Latin, (12) and (13). (10) (11) (12) (13) numa banaga-nyu. (Dixon 1994:10) father-ABS return-NONFUT Father (S) returned yabu-ngu bura-n. (Dixon 1994:10) numa father-ABS mother-ERG see-NONFUT Mother (A) saw father (O) dominus venit. (Dixon 1994:9) master-NOM come-3RDPERS-SG-PRES the master (S) comes dominus servum audit. (Dixon 1994:9) master-NOM slave-ACC hear-3RDPERS-SG-PRES the master (A) hears the slave (O)

Dixon diagrams the relations for a nominative/accusative system and for an absolutive/ergative system as in Figure 2.
NOM/ACC A
NOMINATIVE

ERG/ABS
ERGATIVE

S
ABSOLUTIVE ACCUSATIVE

LEGEND S intransitive subject A transitive subject O transitive object

Fig. 2: Dixons representation of nominative/ergative typology I will not elaborate on Dixons typology, but take it as reflecting the mainstream view on transitivity/ergativity. There are, however, two important comments on Dixons views: the first is a methodological objection, the second relates to the descriptive inadequacy of his observations. The methodological objection concerns Dixons idea that a category can be set up only if there is a reflex in form, which in this case should be either morphological marking or overt syntactic patterning in one syntagm. However, categories are also formally realized in different ways than these two allowed

M. LEMMENS 30 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY by Dixon, viz. by the systematic relations linking all the paradigmatic variants of a construal type. This methodological point has been made by researchers like Whorf (1956) and more recently by Levin (1993). Whorf (1956:87-101) distinguishes overt grammatical categories, or phenotypes, from covert grammatical categories, or cryptotypes. Phenotypes find their realization via a specific formal marking or syntactic patterning. Cryptotypes make themselves felt via their correlation with patterns of syntactic behaviour (cf. also Deane & Wheeler 1984). In Whorfian terms, Dixons typo-logy can thus be criticized as focusing exclusively on the phenotypical level of clause grammar and neglecting the cryptotypical level. Cryptotypes are difficult to define and can only be arrived at by examining their different exponents systematically. Only through systematically tracing these exponents, viz. the various interactions with other patterns in the language, can one arrive at a proper description of the realization and semantic value of crypto-grammar. Levin (1993) can perhaps be regarded as the first mainstream linguist who systematically twists and wrenches verbal constructions. As such, she replaces all the ad hoc arguments used in the literature by a systematic descriptive tool, viz. that of verb alternations. She shows that alternations between construals are a crucial heuristic to the semantics of verb meaning (see section 2.2.3 for a critical discussion of Levins approach). The second point of criticism levelled at Dixons description is its inadequacy for English. In his 1979 article, he acknowledged that English has lexical ergativity, but rejects this as inconsequential because it is not based on overt formal realizations. He further conflates transitive and ergative oneparticipant verbs (e.g. die vs. break), a distinction which will be demonstrated to be crucial in the English grammar of processes. In his recent book on ergativity, Dixon openly objects to the use of the term ergativity for English (pertaining to ambitransitive verbs like break, burst etc.), as in his view it is indisputably an accusative language. He says that [t]he use of ergative in this lexical-semantic/causative sense is quite illicit (in addition to being potentially most confusing) in terms of the well-defined use of the term. (Dixon 1994:20). Dixons confusion results, once again, from his exclusively formal view on ergativity. Following Davidse (1991, 1992), this work reveals that ergativity is important to English clause grammar and manifests itself in a variety of ways. Against the background of the above discussion, the following sections introduce the vast complexities of the transitive and the ergative systems operative in the English grammar of actions and events. The description is organized around four sections. The first one presents Langackers conceptual

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

31

explanation for the nominative/accusative and ergative/absolutive systems. The second section takes the distinction to a more systematic level, discussing Levins (1993) categorization of verbs based on the type of alternations they allow (although Levin herself does not distinguish between ergative and transitive). The third section discusses Davidses (1991) research. She synthesizes many of the relevant insights from other studies, but she regards the issue of transitivity versus ergativity as one of clause grammar and the semantics of construal types. The last section presents my own lexical-paradigmatic position as a further refinement to Davidses framework, drawing on insights from my analysis of verbs of killing. 2.2.2 Langackers cognitive approach Langacker largely follows the traditional typological definition of the nominative/accusative and ergative/absolutive systems. He defines the former (henceforth called the transitive system) as one in which intransitive and transitive subjects have the same form, and the direct object is marked differently (1991b:378). In contrast, an ergative/absolutive system (henceforth, the ergative system) is one in which a transitive subject is specially marked, whereas an intransitive subject has the same form as a transitive object (1991b:379). For languages with case marking, we thus have the following schematization (taken over from Davidse 1991:25):
NOM/ACC NOM NOM verb verb ACC ERG/ABS ABS ERG verb verb ABS

Fig. 3: Typology of nominative and ergative systems Langacker departs from mainstream work on case systems by characterizing case markings as inherently meaningful. Moreover, the transitive and ergative patterns are not only coded by morphological markings, but find numerous other linguistic manifestations (1991b:381). In fact, every language probably uses both patterns in one fashion or another, though the mixture varies and a particular pattern is often predominant (1991b:381).13 From this it follows that

13

Similarly, Halliday observes that probably all transitivity systems, in all languages, are some blend of these two semantic models of processes, the transitive and the ergative (1985:149). Also Dixon (1994) acknowledges that languages mostly show characteristics of both.

M. LEMMENS 32 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY each pattern has a cognitive motivation. But what precisely is the conceptual basis for the transitive/ergative distinction? Langackers description assumes the existence of a cognitive model directly relevant to our conception of actions and events, the billiard-ball model. This archetypal folk model presents our world as being populated by discrete physical objects which, driven by or transmitting energy, are capable of moving through space and making contact with one another (cf. also Talmy 1985a for a similar view in terms of force-dynamics). Transmission of energy may cause the affected object to move or change and possibly to interact with other entities. Such interaction can be represented as an action chain, illustrated as in Figure 4 below (cf. Langacker 1991b:283). The double arrow symbolizes transmission of energy; the change of state which the given object undergoes is represented by the squiggly arrow.

Fig. 4: The action chain Our experience of interacting entities gives rise to yet another model, namely, the stage model. This model idealizes an essential aspect of our ongoing experience: the observation of sequences of external events, each involving the interaction of participants within a setting (Langacker 1991a:210). A sentence like Jill killed Jean with a baseball bat implicitly incor-porates the notion of an external observer who observes Jill as energetically acting upon the bat which in turn acts upon Jean. From our experience of such events, we develop idealized conceptions of typical roles that event participants can assume. In the example sentence, Jill fulfils the role of Agent, the baseball bat that of Instrument and Jean that of Patient. Langacker argues that these role archetypes are pre-linguistic conceptions grounded in everyday experience (1991b:285). The two models combine to form a more complex conceptualization of events, the canonical event model, which can be diagrammed as in Figure 5 (1991b:285).

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

33

AG

INSTR

PAT setting

Fig. 5: Canonical event model Obviously, the canonical event model is a generalization that abstracts from events as coded by particular verbs. As Langacker states it, at the extreme, every verb defines a distinct set of participant roles that reflect its own unique semantic properties (1991b:284). Let us return now to the difference between the transitive and the ergative systems, which Langacker characterizes in terms of the distinct construal paths. The participants marked NOM and ABS (or their equivalents in languages without overt case marking) are usually zero-marked, which indicates that they are the conceptual building blocks of the transitive and ergative systems respect-ively, as is also indicated in Figure 3. The natural construal model for the transitive system is the action chain. The participant marked NOM is head of the energy flow and thereby has the status of primary clausal figure. The participant marked ACC lies downstream and is the secondary clausal figure. Thus, NOM is more basic as it marks the participant that is the starting point of the energy flow. By contrast, in an ergative system, the participant that is conceptually basic (marked ABS) is the secondary clausal figure, i.e. the one that lies downstream from the energy source. Consequently, the conceptual path runs counter to the flow of energy along an action chain. This path is defined in terms of conceptual autonomy/dependence (henceforth A/D) inherent in the structure of conceived events. More specifically, the A/D layering means that in a complex event conception it may be possible to conceptualize certain event components independently of the others, which makes them conceptually autonomous (1991b:286). More specifically, it is thus in principle possible, even for events that are saliently energetic [] to factor out the force-dynamic component and conceive of the change of state autonomously (1991b:289). Talmy (1985b) expresses similar views, saying about the sentence the vase broke that it codes an autonomous event, in the sense that [it] presents an event as occurring in and of itself, without implying that there is a cause (such causes as there may be fall outside of attention) (1985b:79). Figure 6 is an adaptation of how Langacker represents the difference between The glass

M. LEMMENS 34 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY broke, the hammer broke the glass and Floyd broke the glass. The difference concerns first, the profile that is imposed onto the complex event (represented in bold face) and second, the natural path in which the event is assembled (represented by the leftward arrows underneath each box).
(a) The glass broke
scope

(b) The hammer broke the glass

scope

(c) Floyd broke the glass with a hammer

scope

AG setting

INSTR

PAT

AG setting T

INSTR

PAT

AG setting

INSTR

PAT

(E1(T))

(E2(E1(T)))

Fig. 6: Scope, profile and A/D layering of a complex event conception Working outward from the autonomous nucleus, the assembly of a complex event conception implies a mental path which starts with the theme (T) and moves upward towards the energy source (E) of the action chain. This results in event conceptions of progressively greater complexity, represented by Langacker (1991b:382) as [T] > [E1[T]] > [E2[E1[T]]]. The thematic participant is the participant that is evoked as part of the conceptually autonomous core of a processual predication. In the sentences represented in Figure 6, the thematic participant is the glass. The diagram nicely visualizes that the contrast between construal type (b) and (a), which has become known as the causative/noncausative (or inchoative) alternation14, represents a difference of conceptualizations of the event and the two construals are thus semantically non-equivalent. Langacker observes that there are many English verbs, e.g. break, open, melt, starve, burn, freeze, etc. that can be used either transitively or intransitively without any difference in form (1991b:387). These verbs, says Langacker, display an ergative pattern. However, Langacker fails to point out that factoring out the causation component is not always possible. Verbs such as kill, murder, execute, slay, hit, cut, etc. do not allow the non-causative construction. These verbs do allow a formally intransitive construction, as in soldiers going to war must be ready to kill, but this sentence does not mean that the soldiers must prepare to die, but that they must be prepared to be the cause of somebody elses dying. While Langackers characterization of the ergative model in English is partially correct, he offends against his own criterion when he includes in the
14

I will suggest a different terminology in section 2.2.4 which will reflect more closely the paradigmatic oppositions at issue. Until then, I will continue to use the traditional terms causative/non-causative.

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

35

set of ergative patterns a construction like the tomato peels easily. In his opinion, this construction, which is generally called a middle or mediopassive construction, requires its theme to have a skin-like covering (It peeled easily; He peeled it easily) (1991b:388). However, the tomato is not a conceptually autonomous thematic participant as it cannot occur in a noncausative construc-tion: compare *the tomato peeled with the tomato burst. A descriptively more adequate analysis of the middle construction is postponed until Chapter 4, section 4.2. While I do not want to deny the inspirational value of Langackers description, it does not take the empirical observations to their logical conclusion, which would result in a more systematic treatment of the different patterns. The major inaccuracy of Langackers analysis arises from his notion of thematic participant, the participant that stands in a thematic relationship to the process. His characterization of such a thematic relationship as a conceptually autonomous relationship involving just a single participant (1991b:286), leads him to incorrectly identify the single participant in a middle construction as thematic. By characterizing the subject of true intransitives like the tree fell as a thematic participant as well, Langacker conflates under one general denominator grammatical relationships that are clearly semantically distinct. In the case of the following sentences (14) (15) (16) The tree is falling. The window opened. The tomato peels easily.

Langacker attributes to the subject the role of thematic participant and he identi-fies the construal as absolute, i.e. when the autonomous event component is evoked or profiled independently (1991b:389). His notion of absolute construal is too schematic in its alignment of (14) and (15), since for (15) the cause can be incorporated into the clause, e.g. John opened the window, whereas this is not possible for (14), e.g. *John is falling the tree. I argue that this account violates the more subtle distinction in English clause grammar. From Langackers account it follows logically that something that is conceptually dependent cannot be isolated in an absolute construal, as witnessed by his examples. (17)a. The wind caused the tree to fall over. b. The tree fell over. c. *The wind caused. The acceptability of the absolute construal in (17b) is explained by the fact that

M. LEMMENS 36 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY we can [] readily conceptualize this change as occurring autonomously, i.e. in the absence of any external force (1991b:287). The infelicity of (17c) is due to the fact that causation is conceptually dependentthe very notion makes inherent (albeit schematic) reference to the change induced (1991b:287). We will not deny that in many cases an inherent reference to the caused process is present, yet this does not prevent well-formed one-participant structures similar to (17c) to occur. Consider some examples of objectless transitives: (18) (19) These are soldiers trained to kill. Bill always interrupts. (Rice 1988)

The (unfortunate) logical conclusion of Langackers analysis is that the soldiers and Bill are not in a thematic relationship with the process, as he says this relation to be restricted to the subjects of (1) true intransitives (the tree fell), (2) non-causatives (the window opened), and (in my view incorrectly so) (3) middle constructions (the tomato peeled easily). Yet intuitively (and we like to believe that Langacker himself would subscribe to this intuition), the oneparticipant constructions with kill and interrupt are felt to profile what is conceptually basic, focusing on the agents action. Non-causatives, in contrast, focus on the patient (cf. below). To summarize, Langackers account clearly needs a more systematic and all-encompassing analysis which does justice to the subtle semantic distinctions operative in English clause grammar. 2.2.3 Levins lexical-semantic approach In her most inspiring work on English verb classes and what she identifies as diathesis alternations, Levin (1993) goes a long way towards explaining, in a systematic fashion, the alternations in the expression of arguments, sometimes accompanied by changes of meaning (1993:2), in which the English verbs may participate. She is committed to the ideaand I believe correctly sothat speakers of English know which diathesis alternations a verb can have, but also notes that this goes beyond the type of lexical knowledge traditionally represented in subcategorization frames (1993:2) The speakers ability to make judgements about a verbs syntactic behaviour leads Levin to the assumption that general principles of grammar are at work, allowing the syntactic behaviour of a verb to be predicted from its meaning (1993:5). She presents her systematic study of alternations as a heuristic tool towards verb meaning and towards a general classification of verbs. The classification which is most relevant to the present purpose is the distinction between the BREAK verbs and the CUT verbs. The BREAK verbs denote an entity undergoing a change of state (1993:9), and allow the non-

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

37

causative construction (e.g. The cup broke). The CUT verbs, in contrast, do not allow it. The distinction that emerges is thus the following: (20)a. b. (21)a. b. The window broke. (Levin 1993) The little boy broke the window. Margaret cut the string. (Levin 1993) *The string cut. (on the interpretation became cut)

Trying to explain the causative/non-causative alternation with BREAK verbs on semantic grounds, Levin claims that it is found only with verbs of pure change of state (1993:10; my emph.). Although Levins explanation is plausible and applies to a large number of verbs, the notion of pure change of state verbs is too general for the delineation of the categories.15 First, the addition of the qualification pure seems to be drawn into the discussion to remedy the overly general notion of change of state. The BREAK verbs, which comprise verbs such as break, crack, smash, splinter, or tear, are representative examples of the pure change of state verbs allowing the noncausative alternation. By contrast, verbs such as chip, slash, mash, pulverize or squash, which Levin categorizes as CUT verbs, could then be expected to be impure change of state verbs, since they do not allow it. Yet this group undeniably incorporates the notion of a change of state, so much so that in a subsequent categorization Levin herself (1993:241) incorrectly includes chip (a CUT verb) with the BREAK verbs. In general, the CUT verbs do incorporate a more pronounced reference to an agent that causes the change of state, but that does not make them any less change of state verbs. In short, Levins change of state criterion is too broad to capture the necessary distinctions. Secondly, the change of state criterion is linguistically inadequate, since it fails to give an account of why the causative/non-causative alternation is not always applicable to the same verb in different environments (cf. also Davidse 1991:63). For instance, in Levins classification, the verbs change and open are pure change of state verbs allowing the non-causative construction (e.g. John opens the shop at 9:00 a.m. vs. The shop opens at 9:00 a.m.). However, for some clauses with these verbs, there is no corresponding non-causative: (22)a. b. (23)a. b.
15

John opened a tin of baked beans. (Davidse 1991:63) *A tin of baked beans opened. Mother changed the baby. (Davidse 1991:63) *The baby changed (on the interpretation the baby is clean).

Fellbaum (1988:129) quotes Hale & Keyser (1987) as relying on a similar criterion for the class of BREAK verbs and they suggest the rule [x cause [y undergo change], (by)]. She also points out that this rule is a necessary, but not a sufficient condition.

M. LEMMENS 38 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY The constructional variability with the same verb is one of the main issues addressed in this book; I will return to it later in this Chapter, in sections 2.2.4 and 2.2.5. In the final conclusions to this book, the issue will be reconsidered in more theoretical terms. To sum up, Levins change of state criterion fails in two ways: first, it leads to an incorrect classification of verbs and secondly, it ignores the constructional variability that may occur with one and the same verb. Moreover, like Langacker, Levin overlooks the complexity of the middle construction. The middle construction, says Levin, is possible with both BREAK and CUT verbs: (24) (25) Stale bread cuts easily. (Levin 1993) Crystal vases break easily. (Levin 1993)

Once again, this characterization is too general and must be redefined in view of the semantics of both the construction and the verb. As already said, the middle construction is dealt with in Chapter 4, section 4.2, in which I present additional support of my critique on the change-of-state criterion. There are two additional points of criticism regarding Levins theoretical assumptions. First, Levins basic hypothesis is that a verbs syntactic behaviour is predicted by its meaning. While she acknowledges that this hypothesis is not uncontroversial and may eventually turn out to meet with limited success, she maintains the hypothesis, provided, she adds, that one identifies the appropriate representation of verb meaning. Ironically, her own identification of verb meaning that supposedly conditions the difference between CUT and BREAK verbs leads to incorrect predictions and analyses. The meaning of a verb is no doubt pivotal to a proper description of its constructional behaviour yet Levin overrates the extent to which lexical verb meaning determines this behaviour. She relies too much on the semantic properties of verbs, which she sees as independent of the constructions in which they occur (cf. also Goldberg 1995 for similar criticism). For example, objectless constructions like Soldiers trained to kill or You drink too much cannot be explained solely on the basis of the semantics of the verb. In Chapter 5, section 5.4, I argue, amending Rices (1988) lexically-based analysis, that these constructions can be characterized properly only if we take into account the meaning of the grammatical construction itself and the way it modulates the meaning of the verb. Secondly, and this follows logically from the previous point, Levin fails to see that the constructions in which verbs occur are meaningful in and by themselves. Her overgeneralization with respect to the middle construction is due to insufficient attention to the constructional polysemy of this construction (the term constructional polysemy is Goldbergs 1992, 1995). In

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

39

the introduction to her work, Levin says that verb alternations are sometimes accompanied by changes of meaning (1993:2). Against this view, I posit that alternations are inevitably accompanied by changes of meaning since they reflect different ways of construing an event. While Levin sees meaning as independent of clause structure, I hold the view that the clause is where verbal and constructional meaning meet to form a semantic complex capable of triggering unpredictable (yet motivated) creative usages. Levin does not regard such construal variations. This explains why, somewhat surprisingly given her concern with verbal semantics, her distinctions fail to account for cases like (22) and (23), in which verbs that typically belong to a category which allows a certain alternation occur in contexts for which the given alternation no longer applies. In later work, e.g. Levin & Rappaport (1995), a more nuanced view is defended which in many ways come closer to what the present work argues for. A more critical evaluation of Levin & Rappaport (1995) is given in the final conclusion to this work. To conclude this brief description, Levins work can be said to provide an impressive overview of verb classes for which she uses the heuristic tool of a verbs alternations. This allows her to set up a general categorization of verb types. However, her work falls short of recognizing some fundamental general patterns that govern (certain aspects of) English clause grammar. At the same time, she adopts a view on verbal semantics that is too absolute and she consequently overlooks the verbs polysemy which may trigger changes in the type of alternations allowed. The following two sections will address both shortcomings. First, I will consider the general transitive and ergative paradigms at work in the grammar of material processes (section 2.2.4) as described by Davidse (1991, 1992). In the last section (2.2.5), I will characterize how her model should be refined by bringing the semantic flexibility of verbs into the discussion, which is essential in a proper description of English clause grammar. 2.2.4 Davidses paradigmatic approach Davidse (1991) characterizes the grammar of material processes as Janusheaded, i.e. governed by two distinct systems, the ergative and the transitive models of causality.16 These models determine not only the constructional potential of verbs, but also verb-derived grammatical structures, like nominaliz-ation or affixation. I adopt Davidses dictum that
one cannot even begin to describe with any accuracy the grammar of actions and events in English, if one does not take into account the existence of the transitive and
16

See also Davidse (1992, 1994) for additional discussions.

40

M. LEMMENS
LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY ergative systems. But if one does, many long-standing problem-cases in descriptive linguistics can be given a more precise paradigmatic location [] and can also be satisfactorily characterized and explained. (1991:19-20)

In what follows, I will give a general presentation of the transitive and the ergative paradigms, without elaborately reproducing the various exponents (syntactic and/or morphological) which Davidse provides as evidence in support of the workings of the paradigms. I should further note that my concern is with material processes only (the grammar of actions and events observed out there), since the other two large categories of processes which Davidse discusses, i.e. relational processes (the grammar of semiotic relations; e.g. symbolize, resemble, etc.) and mental processes (the grammar of consciousness and its projections; e.g. please, like, look, worry, etc.) are not directly relevant to my purpose. Davidse offers an original synthesis of a number of studies17 by positing transitivity and ergativity as distinct networks of construals. These two semantic models project different process-participant constellations in their conception of events, i.e. different ways in which the participants participate in the process. Halliday (1967-8) talks about differences in the inherent voice relation between the participants and the process. In an ergative two-participant structure, the participant affected by the event is presented as the one most crucially involved; in addition to being affected it also co-participates in the process. Davidse terms this participant the Medium, a term that I will take over (It may be helpful to think of a medium in the ESP sense: it is affected yet at the same time the centre through which the process manifests itself.) It is the Mediums co-participation (and not Levins change of state) which distinguishes an ergative from a transitive two-participant con-strual and determines the semantic difference between the two models of causality encoded. The structural centrality of the MEDIUMPROCESS nucleus is reflected in the fact that it is precisely this nucleus which can be isolated in the non-causative construction, The window broke. The semantic value of the non-causative construction is that it neutralizes whether the process was self-instigated or instigated by an external Instigator. As Smith (1978) puts it, the construction is positively marked for the features of external control as well as independent activity (cf. also Haspelmath 1993:90). However, the ergative system prototypically opens up to the left to incorporate an Instigator, the causer of the action (e.g. The boy broke the window), which

17

The most important of these include various publications by Halliday, Hopper & Thompson (1980), Keyser & Roeper (1984), Rice (1987a,b), and Langacker (1991b).

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

41

resolves the voice vagueness.18 The ergative model is thus one of cause-&effect (Halliday 1985:146) or one of instigation (Davidse 1991:61). This entails that Guerssel et al.s (1985:54) view on the non-causative as having a single passive participant but no active participant is essentially incorrect: non-causative constructions profile the Medium, an entity that is affected by the process, yet is at the same time also an agentive participant. Figure 7, taken over from Davidse (1991), offers a schematic representation of the ergative model. In 7(a) the straight lines denote the conceptual dependency relations: the process is conceptually dependent on the Medium, the entity functioning as the figure in the ergative construal (indicated in bold face).19 The Venn diagram indicates that the PROCESS-MEDIUM cluster is semi-autonomous vis--vis the instigation. In similar vein, Smith (1978:101) talks about the relative independence of the activity. Figure 7(b) visualizes the ergative system as a nuclear one with two processual layers, the instigated process and the instigation of the process, which need not be co-extensive in time or space (Davidse 1991:67ff). The arrow on the inner circle indicates that the process can be self-instigated.
(a) (b)

Is

Process

Me

Is

Me

Fig. 7: The ergative paradigm The nuclear ergative pattern is fundamentally different from the linear transitive paradigm which has an inert Affected. The transitive paradigm takes the Agent as its centre. To distinguish this Agent from the ergative Instigator, Davidse uses the term Actor to designate the category. Both the Instigator and the Actor are instantiations of the more schematic category Agent. The basic building block of the transitive paradigm is thus the ACTOR-PROCESS combination, which can, in some cases, be isolated in objectless transitive constructions, e.g. God saves or Soldiers trained to kill (see Chapter 5, section 5.4). In
18

This term does not cover the same ground as Fillmores use of it; he uses it in a definition of the agentive case, reserved for the typically animate perceived instigator of the action (1968:24). 19 The notion of figure (indicated by bold face) is a cognitive grammar reinterpretation of Davidses figures.

M. LEMMENS 42 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY other words, the transitive system is Actor-centred and right-oriented: the event finds its starting point in the Actor, the most central participant, and in the twoparticipant construction it opens up to the right to include the participant onto which the Actor directs his action, the Goal. The Goal does not co-participate in the process, but is an inert participant wholly affected by the Actors action. As such, the transitive system emphasizes a deed-&-extension aspect of a process (Halliday 1985:146; Davidse 1991:23,61). Figure 8 is how Davidse visualizes the difference from the ergative system: in the transitive system, the ACTOR-PROCESS combination is the nuclear building block (a), and the system is a linear one (b) that may extend to the right.
(a) (b)

Ac

Process

Go

Ac

Go

Fig. 8: The transitive paradigm The transitive and ergative paradigms form two distinct models of causality, each imposing its own image on an event. Since they construe a process in a particular way, the models are inherently meaningful and semantically nonequivalent. In addition, they help explain the specific semantic structure of lexical items that instantiate either of them. Obviously, the schemata do not exhaustively define the semantics of the instantiating clause; our research reveals how verbal semantics as well as other elements in the clause (e.g. particles) make a crucial contribution too. Like most conceptual categories, the ergative and transitive paradigms constitute complex networks of related construals, structured around a prototype. A further discussion of their complex internal structure will be postponed until Chapter 3, section 3.2, following a more elaborate discussion of the prototype-based structure of conceptual categories. In line with the schematic representation of the transitive and ergative case typology (see Figure 3), the situation can be represented as follows (taken over from Davidse 1991:25):

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

Transitive Ergative

43

Ac Ac

proc proc Go

Me Is

proc proc Me

Fig. 9: Semantic reinterpretation of the traditional typology Despite Dixons objections, I subscribe to the view that ergativity is relevant for English, provided that one realizes that it does not manifest itself in overt case markings but in various alternation patterns, which reveal the transitive and ergative models as more covert categories of grammar. As said, the two paradigms make themselves felt in various areas of the grammar. The reader is referred to Davidse (1991) for an elaborate discussion of these exponents; for the present purpose, it suffices to briefly mention some of them as a further clarification. Some of these will be taken up later, when relevant to the analysis of verbs of killing. In view of the above discussion, two-participant constructions are to be distinguished paradigmatically: John killed Mary is a transitive construction, as indicated by the possible alternation with John killed, whereas Bill broke the window is an ergative construction, as indicated by the possible alternation with the window broke. Although formally a one-participant construction, the objectless transitive e.g. John killed is still regarded as a transitive (see Chapter 5, section 5.4). For the one-participant constructions, the paradigmatic opposition is between an ergative construction like the window broke and true intransitives like John died or John is running. The latter are defined as a subtype of the transitive, since they centre around the (prototypically volitional) Actor (see Chapter 4, section 4.3.1, on the relationship between intransitive and ergatives). Process modification and anaphoric reference, already mentioned in the discussion of analytical causatives, can reveal paradigmatic cuts between a transitive construal (with only one processual level) and an ergative one (with two processual layers). In a transitive construal, e.g. he flogged the slave for half an hour, the adverbial modifier can pertain only to the ACTOR-PROCESSGOAL unit; in an ergative construal, e.g. John boiled the water for half an hour (Vendler 1984), it may (but need not always) be limited to just the MEDIUMPROCESS nucleus. The different directionalities of the two paradigms is nicely revealed by reflexive constructions. A transitive reflexive involves the contrast between a non-reflexive and reflexive Goal: John killed himself stands in opposition to

M. LEMMENS 44 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY John killed his neighbour. An ergative reflexive, in contrast, involves the question as to whether or not the Instigator is co-referential with the Medium: the machine stopped itself is not in opposition to The machine stopped the car, but to Lizzy stopped the machine (cf. also Keyser & Roeper 1984:405 on the all by itself-phrase). Also in morphological processes, different paradigmatic cuts can be discerned, as reflected in, for instance, re- prefixation (see Davidse 1991:108ff, reinterpreting observations by Keyser & Roeper 1984), the -able derivation (see Chapter 4, section 4.2.2.3) or the -er suffix deriving agentive nominals from verbs (see Chapter 5, section 5.3). At this point, the traditional terminology can be revised into one that more appropriately captures the different paradigmatic distinctions and commonalities. Davidse uses the term effective for both ergative and transitive twoparticipant constructions and middle to refer to the transitive and ergative single-participant constructions. For the two-participant constructions, I will continue the term effective, which is also used in mainstream linguistic descriptions as the semantic counterpart of grammatical transitivity. For instance, Hopper & Thompson say that transitivity is a property of the entire clause, in which the action is typically EFFECTIVE in some way, as it is carried-over or transferred from an agent to a patient (1980:251). By further specifying the paradigmatic opposition, we end up with the contrastive pair transitive effective versus ergative effective. I will not continue Davidses term middle, which I will restrict to the medio-passive construction, e.g. That tent puts up in anybodys yard (Fellbaum 1985), further discussed in Chapter 4, section 4.2. Instead, I suggest noneffective as a cover term for the one-participant constructions and like the effectives, they can be distinguished paradigmatically: what is traditionally called the intransitive (e.g. John died) is thus the transitive non-effective and the non-causative (or inchoative) will be called the ergative non-effective. At first sight this may seem as an unwarranted terminological inflation. However, the terminology proposed here allows a clear presentation on different levels, which can be presented as follows.

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

CONSTRUCTION EFFECTIVE TRANSITIVE ERGATIVE

45

John killed Mary [AC-PROCESS-GO] John killed [AC-PROCESS-(GO)]

John suffocated Mary [IS-PROCESS-ME]

NON-EFFECTIVE

Mary suffocated [ME-PROCESS] Mary died [AC-PROCESS]

PSEUDO-EFFECTIVE

Mary died a gruesome death [AC-PROCESS-RANGE]

He grew a wart (Davidse 1991) [SETTING-PROCESS-ME]

At the most schematic level, the distinction is between effective, non-effective, and pseudo-effective structures. Pseudo-effectives are so called because one participant (underlined in the examples) is not a true participant. As with the effectives and non-effectives, a paradigmatic refinement can be made, since the area of variability is different for both paradigms: for the transitive pseudoeffectives (called Ranged constructions) there is a pseudo-Goal, for the ergative pseudo-effectives (called Setting constructions), a pseudo-Instigator (see Davidse 1991:115-40). The pseudo-effective constructions are not immediately relevant to the field of killing in present-day English, except perhaps for the cognate object construction (one type of Range) which may occur with die. Also, the diachronic description of starve and related hunger and thirst in Chapter 4, section 4.3.1, briefly discusses Ranged constructions that were possible for the latter two verbs. By using the term non-effective the problems raised by the often used term inchoative (e.g. Marantz 1984, Croft 1991, Levin 1993) can be avoided, as it is often (and originally) used in reference to inceptive verbs, e.g. Latin rubescere begin to be red (cf. also Haspelmath 1993:90, fn.3). In addition, the term fails to bring out the paradigmatic parallel to the transitive noneffective. I also avoid Davidses somewhat idiosyncratic (Hallidayan) use of the term middle. One drawback of the term non-effective is that it stands up against the deeply engrained term intransitive. In response to its high entrenchment, I will mostly use this term in the descriptions. Given that within the transitive paradigm the non-effective constructions stand out much more and are much

M. LEMMENS 46 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY more varied than the ergative non-effective ones, it is reasonable to see the intransitive as a larger subgroup of the transitive model. Parallel to the use of intransitive, one could also to use the term inergative for the ergative non-effective structures, and transitive/ergative for the effectives (Brygida Rudzka-Ostyn, pc). While intuitively appealing, this terminology is bound to be confusing, as inergative may incorrectly be identified with unergative which, in the framework of Relation Grammar (cf. Perlmutter 1978) and Government and Binding (cf. Burzio 1986, Fellbaum & Kegl 1989, Levin 1993, Levin & Rappaport 1995) is used not to refer to ergative non-effectives (as I would have it) but to transitive non-effectives. In these frameworks, ergative verbs are referred to as unaccusatives (hence the Unaccusative Hypothesis). For this reason, I thought it better to set up the distinctions in terms of effectiveness in which, for one-participant constructions, the effectiveness is negated. A last divergence between Davidses terminology and mine concerns her use of transitive absolute to refer to a transitive construction without over object, e.g. Soldiers trained to kill. She stretches Langackers notion of absolute construal, a construal which evokes or profiles the conceptually autonomous core, i.e. the autonomous thematic relationship (cf. above).20 While in earlier work I took over this term, my attention has been drawn to its possibly confusing character, as it resembles too much the term absolutive of traditional typology. Hence, I have resorted to a more common terminology, referring to the construction as an objectless transitive. It should be noted, however, that objectless refers to the formal aspect of the construction; as said, the construction still resorts under the effective constructions as it still has an understood object.21 By way of recapitulation, Table 2 lists some of the different instantiations of the two paradigms and aligns my terminology with Davidses and that of Relational Grammar and Government and Binding (where relevant). To sum up, Davidses work reveals how, in the domain of material processes, the relations between participants and processes are assembled differently. As such, her views are quite compatible with Langackers approach sketched above. Yet, unlike Langacker (and many others), Davidse is not biased towards a single model of transitivity but wakes us up to the many
20

The term absolute has been used in this sense before, e.g. Jespersen (1927) or Feltenius (1977:14). 21 At a point at which it is too late to review the entire manuscript of this book, the term objectless effective was suggested to me as a more accurate cover term, making the effective character of the construction explicit. As said, any reference in this work to an objectless construction should be understood as essentially effective.

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

47

subtle distinctions in what she calls the truly Janus-headed grammar of actions and events.
My terminology Transitive effective non-effective or intransitive pseudo-effective: Ranged Davidse (1991) effective middle or intransitive pseudo-effective: Ranged RG & GB transitive, causative unergative (transitive) Examples John killed Mary. Jill died. He died a slow death. He ran a good race. We prowled the city. John kills. Stale bread cuts easily. unaccusative Jill drowned Jean. transitive, causative unaccusative Jean drowned. (transitive) The system burst a pipe. The window opened only with great difficulty.

objectless transitive transitive absolute middle Ergative effective facility-oriented passive effective

non-effective pseudo-effective: Setting middle

ergative middle or non-ergative pseudo-effective: Setting

Table 2: Terminological overview The above description hardly does justice to the far-reaching implications of Davidses work, yet it suffices as a starting-point for my own data-based investigation, which offers further refinements to her general description. 2.2.5 Present cognitive lexical-paradigmatic approach Even though more traditionally oriented typologists may be uncomfortable with the notion of ergativity being applied to English, a formally nominative/accusative language, I subscribe to Davidses paradigmatic views on the grammar of actions and events. The transitive and ergative systems are cognitively real, also for English, where they manifest themselves not in overt case markings, i.e. phenotypically, but at the cryptotypical level. As Davidse says, English realizes its ergativity as it were by a sleight of hand (1991:147), by using the same verb in both effective and non-effective constructions. At the same time, however, through nominative case marking, the ergative effective

M. LEMMENS 48 LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY realizes a schematic transitive image, as an instantiation of the AG-PROC-AF schema. Transitive constructions, on the other hand, realize phenotypically through case marking only one model, yet do so with considerably more specificity and force than the ergatives. In the final conclusions to this work, we will be in a position to evaluate the double-imposure implied by the ergative model. However, my own analysis of verbs of killing has shown that her predominantly grammatical description needs to be further fine-tuned to the interaction of the grammatical patterns and the semantic structure of the verbs that instantiate them. The constructional potential of a particular verb cannot be stated as absolute (i.e. as either transitive or ergative), but depends on the kind of participants that are involved in the process. So, it is not the verb as such, in tat pur, which evokes the transitive or ergative model, but the entire clause construal. Davidse does acknowledge this when she points out that clauses like in the above examples (22) John opened a tin of baked beans and (23) Mother changed the baby no longer allow the ergative effective. These sentences, Davidse (1991:150) says, realize only the transitive goal-achieving model, as the entities realizing the role of the Medium are no longer entities perceived as capable of sustaining and self-instigating the process. Although Davidse notes this constructional variation, her overall description remains pitched at the grammatical level, i.e. aimed at unravelling the two deep persuasions in the grammar of material processes (1991:146). Slightly oversimplifying, her description gravitates too much to the top-down model, viz. it sees the general models of grammar too much as imposing their structures onto the instantiating verb. However, a unified description of the grammar of processes and events requires the inclusion of a lexical perspective, in Hallidayan terms, the more delicate level of grammar. Accepting the overall validity of the workings of the transitive and ergative paradigms in the grammar of causative constructions, the present work furnishes such a lexical perspective by presenting a corpus-based analysis of verbs of killing. The lexical perspective is described to the fullest in the case studies in Part II. It offers a meticulous analysis of the lexical flexibility (or the absence thereof) of verbs, which has a (direct) impact on the verbs constructional properties. Bringing the lexical perspective into the analysis can, for instance, explain why in certain usages a typically transitive verb occurs in an ergative construction and vice versa. Moreover, as our analysis of the middle construction in Chapter 4 will show, verbal meaning can explain why verbs belonging to a certain paradigm (i.e. the transitive) do not occur in a construction typical of that paradigm (the middle construction). Finally, the

THE SEMANTICS OF CAUSATIVE CONSTRUCTIONS

49

analyses in Part II further substantiate how the ergative/transitive opposition seems to be strongly motivated by our experience of the world. More illustrations of the dynamic interplay between the two paradigms and the complex interaction of constructional and lexical meaning will certainly emerge from the corpus-based analyses, yet before broaching these we should more carefully delineate the supra-lexical category of verbs of killing as well as present our cognitive view on lexical meaning.

M. LEMMENS, LEXICAL PERSPECTIVES ON TRANSITIVITY AND ERGATIVITY,